{ "nodes": [ { "ID": "ai-question", "name": "Who built AI?", "type": "Question", "properties": { "As of": "2026-10-18" }, "description": "What were the milestones that built artificial intelligence, who made them, and how did experts, the public, governments and markets react, from Alan Turing's 1950 paper to October 2026?\n\nArtificial intelligence began as a question before it was a technology. In 1950 Turing asked whether machines could think and proposed the imitation game; in 1956 a summer workshop at Dartmouth College gave the field its name. The first programs proved theorems, held simple conversations and moved a robot, and their makers promised fast progress. When the results fell short, governments cut funding in the 1970s and again at the end of the 1980s, after a boom in expert systems collapsed. These two AI winters left neural networks, the idea of learning from data with layers of artificial neurons, to a small group of researchers.\n\nThe turn came in 2012, when a network trained on two gaming graphics cards won the ImageNet image-recognition contest by a wide margin. Deep learning spread to speech, translation and games, and in 2016 AlphaGo beat one of the world's best Go players. In 2017 Google researchers introduced the Transformer, the architecture behind the large language models that followed; OpenAI's GPT series, Google's BERT and studies of scaling showed that bigger models trained on more data and compute kept getting better.\n\nThe release of ChatGPT on 30 November 2022 turned that research into a mass product and set off a race among technology companies. It brought open letters warning of risk, strikes and copyright lawsuits, the first comprehensive AI law in the European Union, US export controls on AI chips to China, and record funding rounds and market values, with Nvidia, the main maker of AI chips, becoming the world's most valuable company. Reasoning models, first shown by OpenAI's o1 in September 2024, and low-cost rivals such as China's DeepSeek-R1 widened the contest. As of October 2026 the leading labs release new models every few months, governments are still deciding how far to regulate them, and investors are debating whether the spending on AI data centres is a bubble.\n\nhttps://en.wikipedia.org/wiki/History_of_artificial_intelligence\nhttps://en.wikipedia.org/wiki/Timeline_of_artificial_intelligence\nhttps://en.wikipedia.org/wiki/AI_boom", "size": [8, 8], "position": { "x": -293.365, "y": 21.36, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-era-6", "name": "Deep learning revolution", "type": "Era", "properties": { "Era order": 6, "Start date": "2012-09-30", "End date": "2017-06-11", "Ongoing": false }, "description": "The deep learning revolution runs from AlexNet's win in the ImageNet challenge on 30 September 2012 to the day before the Transformer paper appeared in June 2017. AlexNet reached a top-5 error of 15.3 percent, more than 10.8 points better than the runner-up, and its authors sold their company to Google in 2013. In the following years deep networks spread from vision to language and games: word2vec (2013), GANs and attention for translation (2014), the Deep Q-network and Google's purchase of DeepMind, ResNet (2015), the founding of OpenAI in December 2015 and AlphaGo's win over Lee Sedol in March 2016. Big companies began to compete for researchers and compute. The boundaries are a convention; the era name is the usual one in histories.\n\nhttps://en.wikipedia.org/wiki/AlexNet\nhttps://en.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol", "size": [9.5, 3], "position": { "x": -140.715, "y": 74.52, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-era-2", "name": "Early optimism", "type": "Era", "properties": { "Era order": 2, "Start date": "1956-06-18", "End date": "1973-12-31", "Ongoing": false }, "description": "Early optimism covers the years from the Dartmouth workshop of 1956 to the end of 1973. Researchers built the first programs that proved theorems, held conversations and moved a robot, and many predicted fast success: in 1957 Herbert Simon said computers would match human chess players within ten years, which took about forty. Funding came mostly from US military agencies. The period includes the Logic Theorist, the perceptron and its 1958 press coverage, ELIZA (1966) and Shakey (1966 to 1972). It also contains the first warning signs: the 1966 ALPAC report on machine translation, Minsky and Papert's book Perceptrons (1969) and, in 1973, the Lighthill report, which closes the period. The boundaries are a convention taken from the standard histories.\n\nhttps://en.wikipedia.org/wiki/Dartmouth_workshop\nhttps://en.wikipedia.org/wiki/Herbert_A._Simon\nhttps://en.wikipedia.org/wiki/AI_winter", "size": [9.5, 3], "position": { "x": -271.915, "y": 74.52, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-era-4", "name": "Expert systems boom", "type": "Era", "properties": { "Era order": 4, "Start date": "1980-01-01", "End date": "1987-12-31", "Ongoing": false }, "description": "The expert systems boom of the 1980s was the period in which rule-based programs that encode a human specialist's knowledge became the commercial face of artificial intelligence. Digital Equipment Corporation first used XCON, written by John McDermott of Carnegie Mellon, in 1980 to configure computer orders; by the mid-1980s it held about 2,500 rules and was estimated to save the company 25 million dollars a year. Governments responded with large programmes: Japan launched its Fifth Generation project in 1982 and the US followed with DARPA's Strategic Computing Initiative in 1983. Specialised Lisp machines from Symbolics and others sold to AI labs. The boom ends in 1987, when cheaper workstations undercut the Lisp machine market. Boundaries are a convention.\n\nhttps://en.wikipedia.org/wiki/Xcon\nhttps://en.wikipedia.org/wiki/Fifth_generation_computer\nhttps://en.wikipedia.org/wiki/Strategic_Computing_Initiative", "size": [9.5, 3], "position": { "x": -228.876, "y": 74.52, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-era-3", "name": "First AI winter", "type": "Era", "properties": { "Era order": 3, "Start date": "1974-01-01", "End date": "1979-12-31", "Ongoing": false }, "description": "The first AI winter is the period from 1974 to 1979 in which funding and confidence in artificial intelligence fell. Histories point to several causes: the 1966 ALPAC report found machine translation 'more expensive, less accurate and slower than human translation', Minsky and Papert's 1969 book Perceptrons stressed the limits of single-layer networks, the 1973 Lighthill report attacked AI's failure to meet its 'grandiose objectives' and led the UK to end support in most universities, and in 1974 DARPA cancelled a speech-understanding contract of about three million dollars a year at Carnegie Mellon after feeling 'duped'. The historian Thomas Haigh argues that no first winter happened, noting that membership of the ACM's AI interest group grew from 1,241 in 1973 to about 3,500 by mid-1978. The boundaries are a convention.\n\nhttps://en.wikipedia.org/wiki/AI_winter", "size": [9.5, 3], "position": { "x": -239.134, "y": 74.52, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-era-1", "name": "Foundations", "type": "Era", "properties": { "Era order": 1, "Start date": "1950-10-01", "End date": "1956-06-17", "Ongoing": false }, "description": "Foundations is the opening period of the history of artificial intelligence. It starts with Alan Turing's paper 'Computing Machinery and Intelligence' in the October 1950 issue of the journal Mind, which replaced the question 'Can machines think?' with the imitation game, and it ends just before the Dartmouth summer project of 1956 gave the field its name. In these years the questions were set before any program could answer them: Claude Shannon published a paper on programming a computer to play chess and built an electromechanical mouse that learned a maze by trial and error (both 1950), and Marvin Minsky's SNARC machine of 1951 is often called the first artificial neural network. The period boundaries are a convention taken from the standard histories, not a fact of nature.\n\nhttps://en.wikipedia.org/wiki/Computing_Machinery_and_Intelligence\nhttps://en.wikipedia.org/wiki/Claude_Shannon\nhttps://en.wikipedia.org/wiki/Marvin_Minsky", "size": [9.5, 3], "position": { "x": -282.015, "y": 74.52, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-era-8", "name": "Generative AI boom", "type": "Era", "properties": { "Era order": 8, "Start date": "2022-11-30", "End date": "2024-09-11", "Ongoing": false }, "description": "The generative AI boom is the period that opens with the public launch of ChatGPT on 30 November 2022 and runs to the release of OpenAI's o1 reasoning model in September 2024. In those twenty-one months chat assistants reached hundreds of millions of people, Microsoft, Google, Meta and Amazon rushed rival products and investments to market, open letters and hearings put AI risk on government agendas, the first lawsuits and strikes over training data and jobs began, and the EU, the United States, the United Kingdom and China wrote their first binding or declaratory rules. The boundaries are a convention used here, not an official dating: the start follows the first mass-market chatbot, the end follows the point at which commentators began to describe reasoning models and the compute race as a new phase.\n\nhttps://en.wikipedia.org/wiki/ChatGPT\nhttps://en.wikipedia.org/wiki/AI_boom", "size": [9.5, 3], "position": { "x": 47.985, "y": 74.52, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-era-9", "name": "Reasoning and compute", "type": "Era", "properties": { "Era order": 9, "Start date": "2024-09-12", "End date": "2026-10-18", "Ongoing": true }, "description": "Reasoning models and the compute race is a working name for the period that begins with OpenAI's o1 preview on 12 September 2024 and continues to October 2026. No standard name exists yet. In this period labs shifted from simply larger pretraining to models that spend extra computation thinking before they answer (o1, DeepSeek-R1, later GPT-5 and Claude Opus and Mythos lines), companies announced multi-gigawatt data-centre commitments such as Stargate, DeepSeek's January 2025 release triggered a record one-day market loss for Nvidia, US policy swung from Biden's executive order to deregulation and then to direct state intervention in model access, and the first IPO filings by OpenAI and Anthropic appeared in 2026. The start date is a convention chosen for the first widely covered reasoning model.\n\nhttps://en.wikipedia.org/wiki/OpenAI_o1\nhttps://en.wikipedia.org/wiki/2025_in_artificial_intelligence\nhttps://en.wikipedia.org/wiki/2026_in_artificial_intelligence", "size": [9.5, 3], "position": { "x": 164.573, "y": 74.52, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-era-5", "name": "Second AI winter", "type": "Era", "properties": { "Era order": 5, "Start date": "1988-01-01", "End date": "2012-09-29", "Ongoing": false }, "description": "Full name: Second AI winter and statistical learning. From 1988 to September 2012 AI fell out of public favour, then quietly rebuilt itself on statistics and data. The second AI winter began with the collapse of the Lisp machine market in 1987, which Wikipedia's AI winter article says replaced an industry 'worth half a billion dollars' in a single year, followed by deep cuts to DARPA's Strategic Computing programme in 1987 and 1988 and by expert systems that proved brittle and costly to maintain. Japan's Fifth Generation project ended without a breakthrough. Meanwhile neural networks and learning methods advanced out of the spotlight: LeNet read handwritten digits, LSTM appeared in 1997, IBM's Deep Blue beat Garry Kasparov in 1997, Stanley won the DARPA Grand Challenge in 2005, ImageNet was published in 2009 and IBM's Watson won on Jeopardy! in 2011. The period ends the day before AlexNet's ImageNet result. Boundaries are a convention.\n\nhttps://en.wikipedia.org/wiki/AI_winter\nhttps://en.wikipedia.org/wiki/Fifth_generation_computer", "size": [9.5, 3], "position": { "x": -207.738, "y": 74.52, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-era-7", "name": "Transformers and scale", "type": "Era", "properties": { "Era order": 7, "Start date": "2017-06-12", "End date": "2022-11-29", "Ongoing": false }, "description": "The era of transformers and scale runs from the publication of 'Attention Is All You Need' on 12 June 2017 to 29 November 2022, the day before ChatGPT was released. In these years research converged on one architecture, the transformer, and on one recipe: pretrain a very large model on web-scale text or images, then adapt it with prompts or human feedback. Scaling-law studies (2020) and compute-optimal training (2022) gave the recipe numbers, Nvidia GPUs and Google TPUs supplied the compute, and Microsoft's 2019 investment in OpenAI paid for it. Public products such as the GPT-3 API, GitHub Copilot, DALL-E 2, Midjourney and Stable Diffusion followed, and with them the first rules aimed at AI and at the chips that train it: China's 2017 plan, the EU proposal of 2021 and the US export controls of October 2022. The start and end dates are a convention of this history, not an official periodisation.\n\nhttps://en.wikipedia.org/wiki/Attention_Is_All_You_Need\nhttps://arxiv.org/abs/2001.08361", "size": [9.5, 3], "position": { "x": -64.89, "y": 74.52, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-turing-1950", "name": "Turing's 1950 paper", "type": "Paper", "properties": { "Field": "Symbolic AI", "Paper kind": "Theory or position", "Venue": "Mind", "Date": "1950-10-01", "Milestone": true, "Impact score": 75, "URL": "https://doi.org/10.1093/mind/LIX.236.433" }, "description": "Computing Machinery and Intelligence is the paper in which Alan Turing asked 'Can machines think?'. It appeared in the journal Mind (volume 59, number 236, pages 433 to 460) in October 1950. Turing argued that 'think' and 'machine' are too vague to settle the question and replaced it with the imitation game, now called the Turing test: a judge converses by typing with a computer and a human and must decide which is which. The paper answers nine objections to machine intelligence, among them religious, mathematical and consciousness arguments and Lady Lovelace's objection, and proposes programming a child-like machine and educating it with reward and punishment. It became the central text of the philosophy of AI; the test is still used as a reference point and is also widely criticised, for example in John Searle's Chinese room argument.\n\nhttps://doi.org/10.1093/mind/LIX.236.433\nhttps://en.wikipedia.org/wiki/Computing_Machinery_and_Intelligence", "size": [9.8, 2.89], "position": { "x": -286.765, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-pa-perceptrons-book", "name": "Perceptrons book", "type": "Paper", "properties": { "Field": "Neural networks", "Paper kind": "Theory or position", "Venue": "MIT Press book", "Date": "1969-01-01", "Milestone": false, "Impact score": 50, "URL": "https://en.wikipedia.org/wiki/Perceptrons_(book)" }, "description": "Perceptrons: An Introduction to Computational Geometry is the 1969 book by Marvin Minsky and Seymour Papert (MIT Press) that analysed what single-layer perceptrons can and cannot compute. The authors proved limits such as the inability to compute parity under local connectivity, and the book is dedicated to Frank Rosenblatt. It is often thought to have caused the decline of neural-network research in the 1970s and early 1980s and a shift to symbolic AI; critics say the story is flawed, because the book did not claim that multilayer networks share the limits and networks of such units can compute any Boolean function. The authors themselves argued that perceptron research waned because of inherent problems, not because of their book. An expanded edition with a prologue and epilogue appeared in 1988.\n\nhttps://en.wikipedia.org/wiki/Perceptrons_(book)\nhttps://en.wikipedia.org/wiki/Perceptron", "size": [9.8, 2.5], "position": { "x": -254.582, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-pa-hopfield-1982", "name": "Hopfield network", "type": "Paper", "properties": { "Field": "Neural networks", "Paper kind": "Architecture", "Venue": "Proceedings of the National Academy of Sciences", "Date": "1982-04-01", "Milestone": true, "Impact score": 70, "URL": "https://www.pnas.org/doi/10.1073/pnas.79.8.2554" }, "description": "Full title: 'Neural networks and physical systems with emergent collective computational abilities'. In April 1982 the physicist John Hopfield published this paper in the Proceedings of the National Academy of Sciences. It describes a network of binary threshold units that works as an associative memory: patterns are stored with a Hebbian rule and recalled by letting the network settle into a stable state. Hopfield defined an energy function that never increases as units update, so the network always converges to a local minimum, and he used methods from the physics of spin glasses to analyse it. A 1984 follow-up extended the model to continuous activations. The Hopfield network revived interest in neural networks after the years following Perceptrons, and in 2024 the Nobel Committee cited Hopfield and Geoffrey Hinton 'for foundational discoveries and inventions that enable machine learning with artificial neural networks'.\n\nhttps://www.pnas.org/doi/10.1073/pnas.79.8.2554\nhttps://en.wikipedia.org/wiki/Hopfield_network\nhttps://en.wikipedia.org/wiki/John_Hopfield", "size": [9.8, 2.82], "position": { "x": -226.654, "y": 44.15, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-backprop-1986", "name": "Backpropagation", "type": "Paper", "properties": { "Field": "Neural networks", "Paper kind": "Training method", "Venue": "Nature", "Date": "1986-10-09", "Milestone": true, "Impact score": 85, "URL": "https://doi.org/10.1038/323533a0" }, "description": "Full title: 'Learning representations by back-propagating errors'. This three-author paper by David Rumelhart, Geoffrey Hinton and Ronald Williams appeared in Nature on 9 October 1986 and showed that the back-propagation procedure can train multi-layer neural networks to develop useful internal representations. The idea of propagating errors backwards had earlier roots: the chain rule goes back to Leibniz, optimal-control theory of the 1950s contained the adjoint method, and Frank Rosenblatt used the term 'back-propagating error correction' in 1962 without being able to implement it. The 1986 paper, together with the two-volume Parallel Distributed Processing books of the same year, popularised the method and ended the belief, left over from Minsky and Papert's Perceptrons, that deep networks could not be trained. Backpropagation remains the training algorithm of almost all later systems, from LeNet to AlexNet to GPT.\n\nhttps://doi.org/10.1038/323533a0\nhttps://en.wikipedia.org/wiki/Backpropagation\nhttps://en.wikipedia.org/wiki/David_Rumelhart", "size": [9.8, 3.02], "position": { "x": -217.23, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-lstm", "name": "LSTM", "type": "Paper", "properties": { "Field": "Neural networks", "Paper kind": "Architecture", "Venue": "Neural Computation", "Date": "1997-11-15", "Milestone": true, "Impact score": 75, "URL": "https://doi.org/10.1162/neco.1997.9.8.1735" }, "description": "Long Short-Term Memory is the 1997 paper by Sepp Hochreiter and Jürgen Schmidhuber in Neural Computation (volume 9, issue 8, pages 1735 to 1780) that introduced the LSTM recurrent network. Its 'constant error carousel' units address the vanishing gradient problem, which stops ordinary recurrent networks from learning long-range dependencies; Hochreiter had developed the idea in his 1991 diploma thesis. LSTM became the dominant language and speech architecture of the 2010s: Google reported that an LSTM trained with CTC cut transcription errors in Google Voice by 49 percent (2015), Google's neural machine translation system of 2016 used deep LSTMs and reported 60 percent fewer translation errors in a human evaluation, and Apple and Amazon adopted LSTMs for Siri and Polly. Transformers later replaced it in language modelling.\n\nhttps://doi.org/10.1162/neco.1997.9.8.1735\nhttps://en.wikipedia.org/wiki/Long_short-term_memory\nhttps://arxiv.org/abs/1609.08144\nPicture: diagram, NN LSTM-Cell v2.svg, Leouscin, CC0, https://commons.wikimedia.org/wiki/File:NN_LSTM-Cell_v2.svg", "size": [4.2, 2.89], "position": { "x": -193.889, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["4f6cb093-ec93-45a3-8872-2311b2788ee4"] }, { "ID": "ai-hist-pa-deep-belief-nets", "name": "Deep belief nets", "type": "Paper", "properties": { "Field": "Neural networks", "Paper kind": "Training method", "Venue": "Neural Computation", "Date": "2006-07-01", "Milestone": false, "Impact score": 50, "URL": "https://direct.mit.edu/neco/article/18/7/1527/7065/A-Fast-Learning-Algorithm-for-Deep-Belief-Nets" }, "description": "Full title: 'A fast learning algorithm for deep belief nets'. In 2006 Geoffrey Hinton, Simon Osindero and Yee-Whye Teh published this paper in the journal Neural Computation. It showed that a deep belief network, a stack of restricted Boltzmann machines, can be trained greedily one layer at a time with contrastive divergence, so that each layer learns features of the layer below, and that the result can then be fine-tuned with supervised backpropagation. The observation that deep networks could be trained this way is described as one of the first effective deep learning algorithms. At a time when deep networks were considered hard to train, the paper gave Hinton's group, and later the wider field, a recipe and a reason to try again; it is often named as the start of the modern revival of deep learning.\n\nhttps://en.wikipedia.org/wiki/Deep_belief_network\nhttps://en.wikipedia.org/wiki/Deep_learning\nPicture: diagram, Deep belief net.svg, Qwertyus, CC0, https://commons.wikimedia.org/wiki/File:Deep_belief_net.svg", "size": [9.8, 2.5], "position": { "x": -177.766, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["6584b28d-e28f-40fe-a0b0-3afead251baa"] }, { "ID": "ai-hist-pa-gpu-deep-learning-2009", "name": "GPU deep learning", "type": "Paper", "properties": { "Field": "Compute", "Paper kind": "Training method", "Venue": "ICML 2009", "Date": "2009-06-01", "Milestone": false, "Impact score": 50, "URL": "https://doi.org/10.1145/1553374.1553486" }, "description": "Full title: 'Large-scale deep unsupervised learning using graphics processors'. This 2009 paper by Rajat Raina, Anand Madhavan and Andrew Ng of Stanford, presented at the International Conference on Machine Learning (pages 873 to 880), argued that deep belief networks and sparse coding were too slow on CPUs, which limited them to small models, and proposed massively parallel implementations on graphics processors. The authors reported that their deep belief network ran up to 55 times faster than a dual-core CPU version and that training a four-layer network with 100 million parameters fell from several weeks to about one day; Wikipedia's account of the same work gives a figure of up to 70 times. It was an early demonstration that gaming GPUs could make large neural networks practical, three years before AlexNet was trained on two GPUs.\n\nhttps://mlanthology.org/icml/2009/raina2009icml-large\nhttps://doi.org/10.1145/1553374.1553486\nhttps://en.wikipedia.org/wiki/Deep_learning", "size": [9.8, 2.5], "position": { "x": -167.366, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-imagenet", "name": "ImageNet", "type": "Paper", "properties": { "Field": "Vision", "Paper kind": "Dataset or benchmark", "Venue": "CVPR 2009", "Date": "2009-06-20", "Milestone": true, "Impact score": 80, "URL": "https://www.image-net.org/static_files/papers/imagenet_cvpr09.pdf" }, "description": "ImageNet is the image database built by Fei-Fei Li and collaborators at Princeton and first presented at the CVPR conference in 2009, in the paper 'ImageNet: A Large-Scale Hierarchical Image Database' by Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li and Li Fei-Fei. The paper reported 3.2 million labelled images in 5,247 categories (synsets) spread over 12 subtrees. Li began the project in 2007 while at Princeton (another Wikipedia page says 2006); labels were collected from workers on Amazon Mechanical Turk between July 2008 and April 2010, and the database later grew to more than 14 million images in over 20,000 categories. From 2010 an annual contest, the ImageNet Large Scale Visual Recognition Challenge, used a subset with 1,000 classes. The challenge became the benchmark on which deep learning proved itself, when AlexNet won in 2012; the organisers said that the 2017 competition would be the last, since the benchmark was considered solved.\n\nhttps://en.wikipedia.org/wiki/ImageNet\nhttps://www.image-net.org/static_files/papers/imagenet_cvpr09.pdf\nhttps://en.wikipedia.org/wiki/Fei-Fei_Li", "size": [7.4, 2.96], "position": { "x": -158.166, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-pa-dropout", "name": "Dropout", "type": "Paper", "properties": { "Field": "Neural networks", "Paper kind": "Training method", "Venue": "arXiv preprint", "Date": "2012-07-03", "Milestone": false, "Impact score": 45, "URL": "https://arxiv.org/abs/1207.0580" }, "description": "Full title: 'Improving neural networks by preventing co-adaptation of feature detectors'. This July 2012 preprint by Geoffrey Hinton, Nitish Srivastava, Alex Krizhevsky, Ilya Sutskever and Ruslan Salakhutdinov introduced dropout. The authors observed that large feedforward networks trained on small datasets overfit and perform poorly on held-out data, and showed that randomly omitting half of the feature detectors on each training case greatly reduces the problem, because no detector can rely on the presence of specific others and each must learn features that are useful in many contexts. They reported large gains on many benchmark tasks and said dropout 'sets new records for speech and object recognition'. Krizhevsky and Sutskever used the technique in AlexNet a few months later, and it became a standard regulariser of deep networks.\n\nhttps://arxiv.org/abs/1207.0580", "size": [6.6, 2.41], "position": { "x": -148.754, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-pa-word2vec", "name": "word2vec", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Architecture", "Venue": "arXiv preprint", "Date": "2013-01-16", "Milestone": false, "Impact score": 50, "URL": "https://arxiv.org/abs/1301.3781" }, "description": "Full title: 'Efficient Estimation of Word Representations in Vector Space'. Posted to arXiv on 16 January 2013 by Tomas Mikolov, Kai Chen, Greg Corrado and Jeffrey Dean of Google, this paper proposed two architectures, continuous bag-of-words and skip-gram, for learning continuous vector representations of words from very large datasets. The authors reported large accuracy gains at much lower computational cost and stated that it 'takes less than a day to learn high quality word vectors from a 1.6 billion words data set'. A second paper at NeurIPS 2013 added negative sampling and phrase representations and won the 2023 NeurIPS Test of Time Award. Word2vec made it practical to learn embeddings from very large corpora, which contributed to widespread adoption in language processing; by 2022 it was described as dated, as contextual models such as ELMo and BERT replaced static embeddings.\n\nhttps://arxiv.org/abs/1301.3781\nhttps://en.wikipedia.org/wiki/Word2vec", "size": [7.4, 2.5], "position": { "x": -140.377, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-pa-rnn-encoder-decoder", "name": "RNN encoder-decoder", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Architecture", "Venue": "EMNLP 2014", "Date": "2014-06-03", "Milestone": false, "Impact score": 45, "URL": "https://arxiv.org/abs/1406.1078" }, "description": "Full title: 'Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation'. Posted to arXiv on 3 June 2014 and published at EMNLP 2014, this paper by Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk and Yoshua Bengio introduced the RNN Encoder-Decoder: one recurrent network maps an input sequence to a fixed-length vector and a second maps that vector to an output sequence, with both trained jointly to maximise the probability of the target sequence given the source. Adding the model's phrase-pair probabilities as a feature improved a standard statistical translation system. With the Google paper by Sutskever, Vinyals and Le it is cited as one of the two origins of sequence-to-sequence learning.\n\nhttps://arxiv.org/abs/1406.1078\nhttps://en.wikipedia.org/wiki/Seq2seq", "size": [9.8, 2.41], "position": { "x": -124.065, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-gans", "name": "GANs", "type": "Paper", "properties": { "Field": "Generative media", "Paper kind": "Architecture", "Venue": "NIPS 2014", "Date": "2014-06-10", "Milestone": true, "Impact score": 65, "URL": "https://arxiv.org/abs/1406.2661" }, "description": "Generative Adversarial Networks was posted to arXiv on 10 June 2014 by Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville and Yoshua Bengio of the Université de Montréal and presented at NIPS 2014. It proposed training two models at once: a generator that learns the data distribution and a discriminator that estimates whether a sample came from the training data or from the generator, in a minimax game in which the generator tries to maximise the discriminator's error. When both are multilayer perceptrons the whole system can be trained with backpropagation. The paper opened the field of generative modelling with neural networks; GAN variants such as DCGAN and StyleGAN later produced photorealistic images, and the technique was widely used for synthetic faces.\n\nhttps://arxiv.org/abs/1406.2661\nhttps://en.wikipedia.org/wiki/Generative_adversarial_network\nPicture: diagram, Generative adversarial network.svg, Zhang, Aston and Lipton, Zachary C. and Li, Mu and Smola, Alexander J., CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Generative_adversarial_network.svg", "size": [4.2, 2.75], "position": { "x": -122.523, "y": 44.15, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["0ae840e0-1c22-4142-8f0a-a1030f68d193"] }, { "ID": "ai-a-bahdanau-attention", "name": "Bahdanau attention", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Architecture", "Venue": "ICLR 2015", "Date": "2014-09-01", "Milestone": true, "Impact score": 70, "URL": "https://arxiv.org/abs/1409.0473" }, "description": "Full title: 'Neural Machine Translation by Jointly Learning to Align and Translate'. Posted to arXiv on 1 September 2014 by Dzmitry Bahdanau, Kyunghyun Cho and Yoshua Bengio and presented as an oral at ICLR 2015, this paper argued that encoding a whole source sentence into one fixed-length vector limits translation quality. The authors let the model softly search for the parts of the source sentence relevant to each target word, without segmenting them explicitly, and reported that on English-to-French translation the approach was comparable to the best phrase-based system, with soft alignments that agreed with intuition. This 'attention' mechanism was later extended, notably by Luong and colleagues in 2015, and in 2017 the Transformer paper replaced recurrence altogether with self-attention.\n\nhttps://arxiv.org/abs/1409.0473\nhttps://en.wikipedia.org/wiki/Attention_(machine_learning)\nPicture: diagram, Attention mechanism overview.svg, Zhang, Aston and Lipton, Zachary C. and Li, Mu and Smola, Alexander J., CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Attention_mechanism_overview.svg", "size": [9.8, 2.82], "position": { "x": -113.665, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["9e17cf32-1843-41d7-a23a-8e04979680ba"] }, { "ID": "ai-hist-pa-seq2seq", "name": "Seq2seq", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Architecture", "Venue": "NIPS 2014", "Date": "2014-09-10", "Milestone": true, "Impact score": 65, "URL": "https://arxiv.org/abs/1409.3215" }, "description": "Full title: 'Sequence to Sequence Learning with Neural Networks'. Posted to arXiv on 10 September 2014 by Ilya Sutskever, Oriol Vinyals and Quoc Le of Google, this paper used a multilayer LSTM to read a source sentence into a fixed-size vector and a second LSTM to decode the translation. On the WMT'14 English-to-French task the LSTM's translations reached a BLEU score of 34.8 against 33.3 for a phrase-based system, and reranking that system's 1,000 best hypotheses with the LSTM raised the score to 36.5; reversing the word order of source sentences markedly improved performance. The work helped move Google Translate to neural machine translation in 2016. Tomas Mikolov has said he proposed a similar idea earlier and told Sutskever and Le, who did not acknowledge him in the paper, according to Wikipedia.\n\nhttps://arxiv.org/abs/1409.3215\nhttps://en.wikipedia.org/wiki/Seq2seq", "size": [6.6, 2.75], "position": { "x": -104.865, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-pa-nature-deep-learning", "name": "Deep learning review", "type": "Paper", "properties": { "Field": "Neural networks", "Paper kind": "Survey", "Venue": "Nature", "Date": "2015-05-28", "Milestone": false, "Impact score": 40, "URL": "https://doi.org/10.1038/nature14539" }, "description": "Deep learning is the review article by Yann LeCun, Bengio and Geoffrey Hinton published in Nature on 28 May 2015 (volume 521, pages 436 to 444). It explains that deep learning lets computational models with multiple processing layers learn representations of data at multiple levels of abstraction, that these methods had dramatically improved speech recognition, visual object recognition, object detection and fields such as drug discovery and genomics, and that the backpropagation algorithm tells a machine how to change its internal parameters. It describes deep convolutional nets as bringing breakthroughs in images, video, speech and audio and recurrent nets as shedding light on sequential data such as text and speech. The review became the standard citation for the field and was written by the three researchers who shared the 2018 Turing Award.\n\nhttps://doi.org/10.1038/nature14539\nhttps://www.cs.toronto.edu/~hinton/absps/NatureDeepReview.pdf", "size": [9.8, 2.3], "position": { "x": -108.036, "y": 44.15, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-resnet", "name": "ResNet", "type": "Paper", "properties": { "Field": "Vision", "Paper kind": "Architecture", "Venue": "CVPR 2016", "Date": "2015-12-10", "Milestone": true, "Impact score": 75, "URL": "https://arxiv.org/abs/1512.03385" }, "description": "Full title: 'Deep Residual Learning for Image Recognition'. Posted to arXiv on 10 December 2015 by Kaiming He, Xiangyu Zhang, Shaoqing Ren and Jian Sun and presented at CVPR 2016, this paper introduced residual networks, in which layers learn a correction added to their input through shortcut connections. The authors evaluated networks up to 152 layers deep on ImageNet, which they described as eight times deeper than VGG nets with lower complexity; an ensemble reached 3.57 percent error on the ImageNet test set and won first place in the ILSVRC 2015 classification task. Residual connections made very deep networks trainable and became a standard ingredient of later vision networks and, with attention, of translation systems such as Google's GNMT.\n\nhttps://arxiv.org/abs/1512.03385\nhttps://en.wikipedia.org/wiki/Residual_neural_network\nPicture: figure, ResNetBlock.png, Kaiming He Xiangyu Zhang Shaoqing Ren Jian Sun, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:ResNetBlock.png", "size": [5.8, 2.89], "position": { "x": -98.065, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["aae4831b-ded5-431d-acc7-2d09f0214fdc"] }, { "ID": "ai-a-transformer", "name": "Transformer", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Architecture", "Venue": "NeurIPS 2017", "Date": "2017-06-12", "Milestone": true, "Impact score": 85, "URL": "https://arxiv.org/abs/1706.03762" }, "description": "'Attention Is All You Need' is the research paper that introduced the transformer, a neural network built on self-attention instead of recurrence or convolution. Eight researchers at Google (Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan Gomez, Lukasz Kaiser and Illia Polosukhin) posted it on 12 June 2017 and presented it at the NIPS conference that December; the paper lists them as equal contributors in randomised order. The authors argued that attention alone could translate faster and cheaper than the recurrent models then in use, and reported that their base model trained in 12 hours and their big model in 3.5 days on eight Nvidia P100 GPUs. Because the design parallelised well on GPUs and TPUs, it became the basis for BERT, the GPT series, vision transformers and later image and video generators.\n\nhttps://arxiv.org/abs/1706.03762\nhttps://proceedings.neurips.cc/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html\nhttps://en.wikipedia.org/wiki/Attention_Is_All_You_Need\nPicture: figure, Attention Is All You Need - Encoder-decoder Architecture.png, Google, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Attention_Is_All_You_Need_-_Encoder-decoder_Architecture.png", "size": [9.8, 3.02], "position": { "x": -69.64, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["f0aa5636-faf8-4906-a002-5c257e522846"] }, { "ID": "ai-scale-pa-human-preferences", "name": "Human-preference RL", "type": "Paper", "properties": { "Field": "Agents and reasoning", "Paper kind": "Training method", "Venue": "arXiv preprint", "Date": "2017-06-12", "Milestone": false, "Impact score": 48, "URL": "https://arxiv.org/abs/1706.03741" }, "description": "Full title: 'Deep reinforcement learning from human preferences'. This paper, posted on 12 June 2017 by Paul Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg and Dario Amodei, showed that a reinforcement-learning agent could be trained from a person's comparisons between pairs of short clips of its behaviour, with no hand-written reward function. The authors reported that this solved complex tasks including Atari games and simulated robot locomotion, needed human feedback on under 1% of the agent's interactions, and taught novel behaviours with about an hour of human time. It is the origin of reinforcement learning from human feedback, the method that OpenAI later applied to language models in InstructGPT.\n\nhttps://arxiv.org/abs/1706.03741\nhttps://arxiv.org/abs/2203.02155", "size": [9.8, 2.46], "position": { "x": -59.24, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-pa-scaling-laws", "name": "Scaling laws", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Scaling study", "Venue": "arXiv preprint", "Date": "2020-01-23", "Milestone": true, "Impact score": 63, "URL": "https://arxiv.org/abs/2001.08361" }, "description": "'Scaling Laws for Neural Language Models' was posted on 23 January 2020 by Jared Kaplan, Sam McCandlish and eight co-authors, including Tom Brown, Alec Radford and Dario Amodei. The authors reported that the loss of a language model falls as a smooth power law in model size, dataset size and training compute across more than seven orders of magnitude, and that network width or depth matter little within a broad range. From this they concluded that the most compute-efficient way to train is to build a very large model and stop well before convergence. The result gave labs a quantitative reason to scale up, and OpenAI's 175-billion-parameter GPT-3 followed four months later; DeepMind's 2022 Chinchilla study later revised the recommended balance between model size and data.\n\nhttps://arxiv.org/abs/2001.08361", "size": [9.8, 2.72], "position": { "x": -27.276, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-pa-ddpm", "name": "Diffusion models", "type": "Paper", "properties": { "Field": "Generative media", "Paper kind": "Architecture", "Venue": "arXiv preprint", "Date": "2020-06-19", "Milestone": false, "Impact score": 53, "URL": "https://arxiv.org/abs/2006.11239" }, "description": "'Denoising Diffusion Probabilistic Models', posted on 19 June 2020 by Jonathan Ho, Ajay Jain and Pieter Abbeel, showed that diffusion models, which generate images by learning to remove noise step by step, could produce high-quality samples. The authors reported an Inception score of 9.46 and an FID of 3.17 on unconditional CIFAR10 and sample quality similar to ProgressiveGAN on 256 by 256 LSUN. Later image generators such as DALL-E 2 and Stable Diffusion are diffusion models of this family.\n\nhttps://arxiv.org/abs/2006.11239", "size": [9.8, 2.56], "position": { "x": -16.876, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-pa-stochastic-parrots", "name": "Stochastic Parrots", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Theory or position", "Venue": "ACM conference proceedings, March 2021", "Date": "2021-03-01", "Milestone": false, "Impact score": 45, "URL": "https://dl.acm.org/doi/pdf/10.1145/3442188.3445922" }, "description": "'On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?' is a paper by Emily Bender, Timnit Gebru, Angelina McMillan-Major and Margaret Mitchell, published by the ACM in March 2021. It argued that very large language models carry environmental and financial costs, are trained on huge, hard-to-audit internet data that can encode abuse and bias, divert research from approaches aimed at real understanding, and can produce convincing text that misleads. Gebru's dispute with Google over the draft in December 2020 made it widely known before publication.\n\nhttps://dl.acm.org/doi/pdf/10.1145/3442188.3445922\nhttps://www.technologyreview.com/2020/12/04/1013294/google-ai-ethics-research-paper-forced-out-timnit-gebru/", "size": [9.8, 2.41], "position": { "x": -0.308, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-pa-latent-diffusion", "name": "Latent diffusion", "type": "Paper", "properties": { "Field": "Generative media", "Paper kind": "Architecture", "Venue": "arXiv preprint", "Date": "2021-12-20", "Milestone": false, "Impact score": 43, "URL": "https://arxiv.org/abs/2112.10752" }, "description": "Full title: 'High-Resolution Image Synthesis with Latent Diffusion Models'. This paper, posted on 20 December 2021 by Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser and Bjorn Ommer, moved diffusion into the latent space of a pretrained autoencoder instead of working on raw pixels. The authors reported that this cut computing costs sharply while preserving detail, added cross-attention so that text or boxes could condition the output, and released code. It is the research behind Stable Diffusion, which Stability AI, with CompVis at LMU Munich and Runway, published in August 2022 as an open-weights model small enough to run on consumer hardware.\n\nhttps://arxiv.org/abs/2112.10752\nhttps://en.wikipedia.org/wiki/Stable_Diffusion", "size": [9.8, 2.37], "position": { "x": 15.216, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-pa-chain-of-thought", "name": "Chain of thought", "type": "Paper", "properties": { "Field": "Agents and reasoning", "Paper kind": "Training method", "Venue": "arXiv preprint", "Date": "2022-01-28", "Milestone": false, "Impact score": 43, "URL": "https://arxiv.org/abs/2201.11903" }, "description": "Full title: 'Chain-of-Thought Prompting Elicits Reasoning in Large Language Models'. This Google paper, posted on 28 January 2022 by Jason Wei, Denny Zhou, Quoc Le and six co-authors, showed that asking a large language model to write out intermediate reasoning steps, with a few worked examples in the prompt, markedly improved its results on arithmetic, commonsense and symbolic reasoning. The authors reported that a 540-billion-parameter model prompted with eight such examples reached state-of-the-art accuracy on the GSM8K maths benchmark, ahead of a fine-tuned GPT-3 with a verifier. The ability appeared only in sufficiently large models and needed no special training.\n\nhttps://arxiv.org/abs/2201.11903", "size": [9.8, 2.37], "position": { "x": 25.616, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-instructgpt", "name": "InstructGPT", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Training method", "Venue": "arXiv preprint", "Date": "2022-03-04", "Milestone": true, "Impact score": 75, "URL": "https://arxiv.org/abs/2203.02155" }, "description": "Full title: 'Training language models to follow instructions with human feedback'. This OpenAI paper, posted on 4 March 2022 by Long Ouyang, Jeff Wu, John Schulman, Jan Leike, Ryan Lowe and others, introduced InstructGPT. The authors fine-tuned GPT-3 first on labeller-written demonstrations and then with reinforcement learning from human feedback, using rankings of model outputs. In human evaluations the 1.3-billion-parameter InstructGPT was preferred to the 175-billion-parameter GPT-3 despite having about 100 times fewer parameters, and it was more truthful and less toxic, though it still made simple mistakes. The method of aligning a model with human preferences became the recipe for ChatGPT, released on 30 November 2022.\n\nhttps://arxiv.org/abs/2203.02155\nhttps://en.wikipedia.org/wiki/ChatGPT", "size": [9.8, 2.89], "position": { "x": 36.016, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-pa-chinchilla", "name": "Chinchilla", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Scaling study", "Venue": "arXiv preprint", "Date": "2022-03-29", "Milestone": false, "Impact score": 43, "URL": "https://arxiv.org/abs/2203.15556" }, "description": "Full title: 'Training Compute-Optimal Large Language Models'. This DeepMind paper, posted on 29 March 2022 by Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch and 19 co-authors, argued that large language models of the time were significantly undertrained. After training over 400 models from 70 million to over 16 billion parameters on 5 to 500 billion tokens, the authors concluded that model size and training tokens should scale together, doubling the data whenever the model doubles. Their 70-billion-parameter Chinchilla used the same compute as the 280-billion-parameter Gopher with four times more data and outperformed Gopher, GPT-3, Jurassic-1 and Megatron-Turing NLG on a wide range of tasks, reaching 67.5% average accuracy on MMLU as the authors reported. The finding shifted labs toward training smaller models on more data.\n\nhttps://arxiv.org/abs/2203.15556", "size": [9, 2.37], "position": { "x": 46.016, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-pa-sparks-agi", "name": "Sparks of AGI", "type": "Paper", "properties": { "Field": "Language", "Paper kind": "Theory or position", "Venue": "arXiv preprint", "Date": "2023-03-22", "Milestone": false, "Impact score": 23, "URL": "https://arxiv.org/abs/2303.12712" }, "description": "Sparks of Artificial General Intelligence: Early experiments with GPT-4 is a Microsoft Research paper led by Sebastien Bubeck with thirteen co-authors, posted on arXiv on 22 March 2023. It reports tests of an early development version of GPT-4 across mathematics, coding, medicine, law and other domains and concludes that the model 'could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system', while also stressing its limitations. The paper was widely quoted in debates over whether large language models show general intelligence, and Wikipedia's GPT-4 article records the claim next to the critics who disputed the reliability of such tests.\n\nhttps://arxiv.org/abs/2303.12712", "size": [9.8, 1.75], "position": { "x": 66.641, "y": 40.63, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-ms-logic-theorist", "name": "Logic Theorist", "type": "Model or system", "properties": { "Field": "Symbolic AI", "System kind": "Symbolic program", "Open weights": false, "Parameters": "not applicable", "Date": "1956-01-01", "Milestone": false, "Impact score": 45 }, "description": "Logic Theorist was a program completed in 1956 by Allen Newell, Herbert Simon and Cliff Shaw at the RAND Corporation, and it is often described as the first artificial intelligence program. It proved 38 of the first 52 theorems in chapter 2 of Whitehead and Russell's Principia Mathematica and found a shorter proof of one of them (theorem 2.85); the Journal of Symbolic Logic reportedly rejected a paper on that proof because a new proof of an elementary theorem was not notable. Newell and Simon presented the program at the Dartmouth workshop in summer 1956, where the reception was lukewarm. Its lasting ideas were reasoning as search, heuristics and list processing, implemented in the language IPL, which influenced Lisp.\n\nhttps://en.wikipedia.org/wiki/Logic_Theorist\nhttps://archive.org/details/bitsavers_randiplP86ineJul56_3534001", "size": 4.06, "position": { "x": -279.97, "y": 31.92, "z": 0 } }, { "ID": "ai-a-perceptron", "name": "Perceptron", "type": "Model or system", "properties": { "Field": "Neural networks", "System kind": "Neural network", "Open weights": false, "Parameters": "not applicable", "Date": "1958-07-07", "Milestone": true, "Impact score": 75 }, "description": "The perceptron was the first trainable neural network machine, invented by Frank Rosenblatt at the Cornell Aeronautical Laboratory. Rosenblatt simulated it on an IBM 704 in 1957, and on 7 July 1958 the US Navy, which funded the work through the Office of Naval Research, staged a press demonstration in which the machine, after about 50 trials, taught itself to tell cards marked on the left from cards marked on the right. A hardware version, the Mark I Perceptron, was assembled between June and December 1959 and first shown publicly on 23 June 1960. The New York Times reported the next day that the Navy expected the device to be able to 'walk, talk, see' and write; critics later called this hype. The perceptron became the reference point for Minsky and Papert's 1969 critique.\n\nhttps://en.wikipedia.org/wiki/Perceptron\nhttps://news.cornell.edu/node/313946\nhttps://www.nytimes.com/1958/07/08/archives/new-navy-device-learns-by-doing-psychologist-shows-embryo-of.html", "size": 5.25, "position": { "x": -272.713, "y": 20.07, "z": 0 }, "images": ["c82fd64f-aea9-4098-8a79-737a1a72c4e4"] }, { "ID": "ai-hist-ms-eliza", "name": "ELIZA", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Symbolic program", "Open weights": false, "Parameters": "not applicable", "Date": "1966-01-01", "Milestone": true, "Impact score": 60 }, "description": "ELIZA was a conversation program written by Joseph Weizenbaum at MIT between 1964 and 1967 and described in the Communications of the ACM in 1966. It ran on a time-sharing system on an IBM 7094 and used keyword pattern matching and substitution to give an illusion of understanding. Its best-known script, DOCTOR, imitated a Rogerian psychotherapist by turning the user's words back as non-directional questions, which let Weizenbaum avoid giving it a database of real-world knowledge. Users often attributed human feelings to it; Weizenbaum was surprised when his secretary asked him to leave the room so that she could talk to the program, and he later wrote of 'powerful delusional thinking in quite normal people'. The tendency is now called the ELIZA effect.\n\nhttps://en.wikipedia.org/wiki/ELIZA\nhttps://doi.org/10.1145/365153.365168\nPicture: screenshot, ELIZA conversation.png, unknown author, Public domain, https://commons.wikimedia.org/wiki/File:ELIZA_conversation.png", "size": 4.71, "position": { "x": -259.5, "y": 31.92, "z": 0 }, "images": ["8e2f1653-bdb0-4e1f-bc33-8a52d81a15d6"] }, { "ID": "ai-hist-ms-shakey", "name": "Shakey the robot", "type": "Model or system", "properties": { "Field": "Robotics and driving", "System kind": "Vehicle or robot", "Open weights": false, "Parameters": "not applicable", "Date": "1966-01-01", "Milestone": true, "Impact score": 50 }, "description": "Shakey was a mobile robot built at SRI International's Artificial Intelligence Center in Menlo Park from about 1966 to 1972, with DARPA funding; Charles Rosen, Nils Nilsson and Peter Hart managed the project. Its wobbly movement gave it the name. Shakey could take a high-level command and break it into smaller actions by itself, using the STRIPS planner, with software mostly in Lisp: told to push a block off a platform, it found a ramp, moved it to the platform, climbed up and pushed the block. It carried a television camera, sonar range finders and bump detectors. The project produced the A* search algorithm, the Hough transform and the visibility graph for path planning. Life magazine called it the 'first electronic person' in 1970, it was inducted into the Robot Hall of Fame in 2004 and received an IEEE Milestone in 2017, and it is displayed at the Computer History Museum.\n\nhttps://en.wikipedia.org/wiki/Shakey_the_robot\nPicture: SRI Shakey robot, 1969, Computer History Museum.jpg, The wub, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:SRI_Shakey_robot,_1969,_Computer_History_Museum.jpg", "size": 4.29, "position": { "x": -262.367, "y": 20.07, "z": 0 }, "images": ["eef7cc71-198a-42bc-bead-873c540be287"] }, { "ID": "ai-hist-ms-mycin", "name": "MYCIN", "type": "Model or system", "properties": { "Field": "Symbolic AI", "System kind": "Expert system", "Open weights": false, "Parameters": "about 600 rules", "Date": "1972-01-01", "Milestone": false, "Impact score": 45 }, "description": "MYCIN was an expert system developed in the early 1970s at Stanford University by Edward Shortliffe as his doctoral work, under Bruce Buchanan, Stanley Cohen and others. About 600 production rules written in Lisp let it identify bacteria that cause severe infections such as bacteremia and meningitis and recommend antibiotics with doses adjusted to body weight. In an evaluation it received a 65 percent acceptability rating, compared with 42.5 to 62.5 percent for five faculty members. It was never used in routine practice; ethical and legal questions about responsibility were raised, but the main barrier was the lack of ways to integrate a stand-alone system that needed typed input. Its legacy was the EMYCIN shell, which separated domain knowledge from the reasoning engine and shaped later rule-based systems.\n\nhttps://en.wikipedia.org/wiki/Mycin\nhttps://doi.org/10.1016/0025-5564(75)90047-4", "size": 4.06, "position": { "x": -249.074, "y": 31.92, "z": 0 } }, { "ID": "ai-hist-ms-xcon", "name": "XCON", "type": "Model or system", "properties": { "Field": "Symbolic AI", "System kind": "Expert system", "Open weights": false, "Parameters": "about 2,500 rules (mid-1980s)", "Date": "1980-01-01", "Milestone": true, "Impact score": 55 }, "description": "XCON, also called R1, was an expert system that John McDermott of Carnegie Mellon University wrote in 1978 for Digital Equipment Corporation to check and configure orders for its VAX computers. DEC first put it into service in 1980 at DEC's plant in Salem, New Hampshire; McDermott's paper appeared in 1980. It grew to about 2,500 rules at its peak in the mid-1980s and 6,200 by 1987, and DEC was estimated in 1986 to be saving 25 million dollars a year with it. By 1987 maintenance was hard, because the difficulty of updating did not grow linearly with the rule base, and XCON was discontinued in the early 1990s. It is the standard example of the commercial success and later limits of expert systems.\n\nhttps://en.wikipedia.org/wiki/Xcon\nhttps://ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/445", "size": 4.51, "position": { "x": -233.626, "y": 31.92, "z": 0 } }, { "ID": "ai-hist-ms-lenet", "name": "LeNet", "type": "Model or system", "properties": { "Field": "Vision", "System kind": "Neural network", "Open weights": false, "Parameters": "about 60 thousand (LeNet-5)", "Date": "1989-01-01", "Milestone": false, "Impact score": 50 }, "description": "LeNet is the family of convolutional neural networks developed by Yann LeCun and colleagues at AT&T Bell Labs. After LeCun joined the Adaptive Systems Research Department in Holmdel, New Jersey in October 1988, the group applied backpropagation to recognise handwritten zip code digits from US Postal Service data in 1989, the prototype that became LeNet-1. The best-known version, LeNet-5, was described in a 1998 paper by LeCun, Léon Bottou, Yoshua Bengio and Patrick Haffner; it has seven layers and about 60,000 trainable parameters. Bell Labs' work led to ATM check-reading systems, and NCR deployed large check-reading machines in bank back offices from June 1996; by 2001 the system was estimated to read 20 million checks a day, about 10 percent of US checks.\n\nhttps://en.wikipedia.org/wiki/LeNet\nhttps://yann.lecun.com/exdb/lenet/index.html\nhttps://doi.org/10.1109/5.726791\nPicture: diagram, LeNet-5 architecture.svg, Zhang, Aston and Lipton, Zachary C. and Li, Mu and Smola, Alexander J., CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:LeNet-5_architecture.svg", "size": 4.29, "position": { "x": -210.273, "y": 31.92, "z": 0 }, "images": ["ee37899a-34ca-40d4-ba23-893d01b319ea"] }, { "ID": "ai-a-deep-blue", "name": "Deep Blue", "type": "Model or system", "properties": { "Field": "Games", "System kind": "Game-playing system", "Open weights": false, "Parameters": "not applicable", "Date": "1997-05-11", "Milestone": true, "Impact score": 75 }, "description": "Deep Blue was the IBM chess supercomputer that defeated world champion Garry Kasparov in a six-game match in New York in May 1997. A first version had lost to Kasparov 4 to 2 in Philadelphia in February 1996, although it won Game 1 of that match, the first game a chess computer won against a reigning champion under classical time controls. In the 1997 rematch, played from 3 to 11 May at the Equitable Center, Deep Blue won 3.5 to 2.5, the first computer win over a reigning champion in a match under tournament regulations; Game 6 on 11 May ended in under twenty moves after a knight sacrifice, with the score level at 2.5 each beforehand. Herbert Simon had predicted in 1957 that a computer would match human chess players within ten years. IBM was reported to have dismantled the machine, but it remained in operation for several years.\n\nhttps://en.wikipedia.org/wiki/Deep_Blue_versus_Garry_Kasparov\nhttps://www.npr.org/2026/02/09/nx-s1-5684150/deep-blue-ibm-garry-kasparov-1996-ai-machine-learning\nPicture: IBM Deep Blue at Computer History Museum (9361685537).jpg, Anton Chiang from Cupertino, CA, USA, CC BY 2.0, https://commons.wikimedia.org/wiki/File:IBM_Deep_Blue_at_Computer_History_Museum_(9361685537).jpg", "size": 5.25, "position": { "x": -196.718, "y": 20.07, "z": 0 }, "images": ["69ac6fe6-bf18-4618-b2d3-1329e5261750"] }, { "ID": "ai-hist-ms-stanley", "name": "Stanley", "type": "Model or system", "properties": { "Field": "Robotics and driving", "System kind": "Vehicle or robot", "Open weights": false, "Parameters": "not applicable", "Date": "2005-10-08", "Milestone": false, "Impact score": 40 }, "description": "Stanley was an autonomous car built by the Stanford Racing Team with Volkswagen's Electronics Research Laboratory and led by Sebastian Thrun, then at the Stanford Artificial Intelligence Lab. It was a modified Volkswagen Touareg carrying five roof-mounted Sick lidar units, a camera, gyroscopes and accelerometers and a wheel odometer, with seven Pentium M computers running Linux and about 100,000 lines of code; obstacle detection used machine learning, and logs of human driving helped reduce errors and improve shadow detection. It won the 2005 DARPA Grand Challenge and the 2 million dollar prize. The team later entered a Passat wagon, Junior, in the 2007 Urban Challenge. Stanley is displayed at the Smithsonian American History Museum.\n\nhttps://en.wikipedia.org/wiki/Stanley_(vehicle)\nhttps://en.wikipedia.org/wiki/DARPA_Grand_Challenge_(2005)", "size": 3.8, "position": { "x": -166.213, "y": 31.92, "z": 0 }, "images": ["f989e98a-f5aa-4625-b15e-21918ee8509d"] }, { "ID": "ai-hist-ms-watson", "name": "IBM Watson", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Game-playing system", "Open weights": false, "Parameters": "not applicable", "Date": "2011-02-14", "Milestone": false, "Impact score": 50 }, "description": "Watson was the question-answering system that IBM built under the DeepQA project, led by principal investigator David Ferrucci, to compete on Jeopardy!. For the match it ran on a cluster of 90 IBM Power 750 servers with 2,880 POWER7 processor threads and 16 terabytes of memory, reaching 80 teraFLOPS on the Linpack benchmark; Tony Pearson estimated the hardware cost at about 3 million dollars. The software used DeepQA and the Apache UIMA framework, written in Java, C++ and Prolog, on Linux. After the 2011 win IBM marketed Watson to hospitals and companies; Watson Health was sold in 2022 for about 1 billion dollars after costing about 4 billion dollars to develop, according to the reports summarised on Wikipedia.\n\nhttps://en.wikipedia.org/wiki/Watson_(computer)\nhttps://www.discovermagazine.com/technology/what-is-champion-ibms-watson-seals-its-jeopardy-victory\nPicture: IBM Watson.PNG, Clockready, CC BY-SA 3.0, https://commons.wikimedia.org/wiki/File:IBM_Watson.PNG", "size": 4.29, "position": { "x": -161.403, "y": 20.07, "z": 0 }, "images": ["176b0757-5669-43c5-95df-d2c224148a70"] }, { "ID": "ai-hist-ms-google-cat-network", "name": "Google cat network", "type": "Model or system", "properties": { "Field": "Vision", "System kind": "Neural network", "Open weights": false, "Parameters": "about 1 billion connections", "Date": "2012-06-25", "Milestone": true, "Impact score": 65 }, "description": "In 2012 researchers at Google, among them Quoc Le, Jeff Dean, Greg Corrado and Andrew Ng, reported a neural network that learned to detect faces and cat faces without being given labels. The network was a nine-layer locally connected sparse autoencoder with about one billion connections, trained on 10 million 200 by 200 pixel images from the Internet, using model parallelism and asynchronous gradient descent on a cluster of 1,000 machines (16,000 cores) for three days. The authors reported that starting from these learned features they reached 15.8 percent accuracy on recognising 20,000 ImageNet categories, a 70 percent relative improvement on the previous best. A New York Times story of June 2012 ran under the headline 'How Many Computers to Identify a Cat? 16,000'. It was the first widely reported result of Google Brain, the project started in 2011 at Google X by Dean, Corrado and Ng.\n\nhttps://arxiv.org/abs/1112.6209\nhttps://en.wikipedia.org/wiki/Google_Brain", "size": 4.9, "position": { "x": -141.993, "y": 20.07, "z": 0 } }, { "ID": "ai-a-alexnet", "name": "AlexNet", "type": "Model or system", "properties": { "Field": "Vision", "System kind": "Neural network", "Open weights": false, "Parameters": "about 60 million", "Date": "2012-09-30", "Milestone": true, "Impact score": 85 }, "description": "AlexNet is the convolutional neural network built by Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton at the University of Toronto that won the ImageNet Large Scale Visual Recognition Challenge in 2012. Entered as SuperVision, it reached a top-5 error of 15.3 percent on the contest's results page (16.4 percent using only the supplied training data) against 26.2 percent for the best other team, ISI of the University of Tokyo. The network has about 60 million parameters and eight layers, five convolutional and three fully connected, and was trained over five to six days for 90 epochs on the 1.2 million training images, split across two Nvidia GTX 580 graphics cards because it did not fit in the memory of one. The authors sold their start-up DNNresearch, with the code, to Google in 2013; the original source code was released under a BSD-2 licence through the Computer History Museum in 2025. The result convinced much of the field that deep learning with GPUs and large data worked.\n\nhttps://image-net.org/challenges/LSVRC/2012/results.html\nhttps://en.wikipedia.org/wiki/AlexNet\nhttps://papers.nips.cc/paper_files/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\nPicture: diagram, AlexNet block diagram.svg, Zhang, Aston and Lipton, Zachary C. and Li, Mu and Smola, Alexander J., CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:AlexNet_block_diagram.svg", "size": 5.57, "position": { "x": -145.465, "y": 31.92, "z": 0 }, "images": ["874f226f-4bca-4968-a6fa-f04a7ed2e733"] }, { "ID": "ai-hist-ms-dqn", "name": "Deep Q-network", "type": "Model or system", "properties": { "Field": "Games", "System kind": "Game-playing system", "Open weights": false, "Parameters": "not disclosed", "Date": "2013-12-19", "Milestone": true, "Impact score": 65 }, "description": "The Deep Q-network, or DQN, is the system with which DeepMind showed that a single neural network could learn to play video games from pixels and a score. A December 2013 preprint by Volodymyr Mnih, Koray Kavukcuoglu, David Silver and colleagues described a convolutional network trained with a variant of Q-learning on seven Atari 2600 games with no change of architecture; it outperformed earlier approaches on six and surpassed a human expert on three. A Nature paper, 'Human-level control through deep reinforcement learning', followed on 26 February 2015 (volume 518, pages 529 to 533), using experience replay and periodically updated target values, and DeepMind patented a deep Q-learning method in 2014. The work was a key reason Google bought DeepMind and the template for AlphaGo's reinforcement learning.\n\nhttps://arxiv.org/abs/1312.5602\nhttps://doi.org/10.1038/nature14236\nhttps://en.wikipedia.org/wiki/Q-learning", "size": 4.9, "position": { "x": -123.995, "y": 31.92, "z": 0 } }, { "ID": "ai-hist-ms-alphago", "name": "AlphaGo", "type": "Model or system", "properties": { "Field": "Games", "System kind": "Game-playing system", "Open weights": false, "Parameters": "not disclosed", "Date": "2016-01-27", "Milestone": false, "Impact score": 60 }, "description": "AlphaGo is the Go-playing program developed by Google DeepMind. It combines Monte Carlo tree search with a policy network, which narrows the moves considered, and a value network, which estimates the winner of a position; both are deep convolutional networks, trained first on about 30 million moves from human experts and then by self-play. In October 2015 a distributed version beat the European champion Fan Hui 5 to 0, the first time a program beat a professional on a full-size board without handicap; the news was held back until the Nature paper 'Mastering the game of Go with deep neural networks and tree search' appeared on 27 January 2016. DeepMind reported that on a single machine AlphaGo won 499 of 500 games against the strongest other Go programs. Wikipedia gives the October 2015 version 1,202 CPUs and 176 GPUs and the Lee Sedol version 1,920 CPUs and 280 GPUs, citing The Economist. In March 2016 it beat Lee Sedol 4 to 1.\n\nhttps://research.google/blog/alphago-mastering-the-ancient-game-of-go-with-machine-learning/\nhttps://en.wikipedia.org/wiki/AlphaGo", "size": 4.71, "position": { "x": -98.636, "y": 31.92, "z": 0 } }, { "ID": "ai-hist-ms-gnmt", "name": "Google neural translation", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Neural network", "Open weights": false, "Parameters": "not disclosed", "Date": "2016-09-26", "Milestone": false, "Impact score": 45 }, "description": "Google's Neural Machine Translation system, GNMT, was described in a paper posted to arXiv on 26 September 2016 by Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc Le and 27 co-authors. It is a deep LSTM network with eight encoder and eight decoder layers, using attention and residual connections, splitting words into shared sub-word units called wordpieces to handle rare words, and low-precision arithmetic at inference for speed. The paper reported results competitive with the state of the art on English-to-French and English-to-German benchmarks and, in a human side-by-side evaluation on isolated simple sentences, an average 60 percent reduction in translation errors against Google's phrase-based production system. Google moved Google Translate to neural translation based on this line of work.\n\nhttps://arxiv.org/abs/1609.08144\nhttps://en.wikipedia.org/wiki/Seq2seq", "size": 4.06, "position": { "x": -80.921, "y": 31.92, "z": 0 } }, { "ID": "ai-scale-ms-alphago-zero", "name": "AlphaGo Zero", "type": "Model or system", "properties": { "Field": "Games", "System kind": "Game-playing system", "Open weights": false, "Parameters": "not disclosed", "Date": "2017-10-19", "Milestone": false, "Impact score": 45 }, "description": "AlphaGo Zero is DeepMind's Go program that learned the game from self-play alone, with no human game records. DeepMind announced it on 18 October 2017 and published it in Nature on 19 October; after three days of training it beat the version of AlphaGo that had defeated Lee Sedol by 100 games to 0, as DeepMind reported. The training setup used 64 GPU workers and 19 CPU parameter servers, with four TPUs for play. Because it needed no human data, the method generalised quickly: within two months DeepMind posted AlphaZero, which applied the same approach to chess and shogi.\n\nhttps://en.wikipedia.org/wiki/AlphaGo_Zero\nhttps://deepmind.com/blog/alphago-zero-learning-scratch/", "size": 4.06, "position": { "x": -62.297, "y": 31.92, "z": 0 } }, { "ID": "ai-scale-ms-alphazero", "name": "AlphaZero", "type": "Model or system", "properties": { "Field": "Games", "System kind": "Game-playing system", "Open weights": false, "Parameters": "not disclosed", "Date": "2017-12-05", "Milestone": true, "Impact score": 60 }, "description": "AlphaZero is DeepMind's general game-playing system, described in a paper posted on 5 December 2017 and published in Science on 7 December 2018. Using the self-play approach of AlphaGo Zero and no opening books or endgame tables, it taught itself chess, shogi and Go. DeepMind reported a 100-game chess match against Stockfish in 2017 that AlphaZero won with 28 wins, 0 losses and 72 draws, and a 1,000-game match in 2018 with 155 wins, 6 losses and 839 draws; critics, including Stockfish's developer Tord Romstad, objected to the time controls and hardware. Garry Kasparov called it a remarkable achievement, even if it should have been expected after AlphaGo. Self-play was run on 5,000 first-generation TPUs, with 64 second-generation TPUs training the networks.\n\nhttps://arxiv.org/abs/1712.01815\nhttps://en.wikipedia.org/wiki/AlphaZero\nhttps://www.technologyreview.com/2017/12/08/147199/alpha-zeros-alien-chess-shows-the-power-and-the-peculiarity-of-ai/", "size": 4.71, "position": { "x": -52.927, "y": 31.92, "z": 0 } }, { "ID": "ai-scale-ms-gpt-1", "name": "GPT-1", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": false, "Parameters": "0.12 billion", "Date": "2018-06-11", "Milestone": false, "Impact score": 38 }, "description": "GPT-1 was OpenAI's first generative pretrained transformer, released on 11 June 2018 with the paper 'Improving Language Understanding by Generative Pre-Training' by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever. It is a 12-layer, 12-headed transformer decoder with about 0.12 billion parameters, pretrained on BookCorpus (about 4.5 GB of text from 7,000 unpublished books) and then fine-tuned for individual tasks. OpenAI's blog post presented the approach as unsupervised pretraining followed by supervised fine-tuning, the pattern that its successors GPT-2 and GPT-3 scaled up.\n\nhttps://openai.com/research/language-unsupervised\nhttps://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf\nhttps://en.wikipedia.org/wiki/Generative_pre-trained_transformer\nPicture: diagram, Full GPT architecture.svg, Original: Marxav, Vectorization: Mrmw, CC0, https://commons.wikimedia.org/wiki/File:Full_GPT_architecture.svg", "size": 3.69, "position": { "x": -43.927, "y": 31.92, "z": 0 }, "images": ["bebbce40-e764-43f6-b9cf-e491bd932ed3"] }, { "ID": "ai-a-bert", "name": "BERT", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": true, "Parameters": "110 million (base), 340 million (large)", "Date": "2018-10-11", "Milestone": true, "Impact score": 60 }, "description": "BERT (Bidirectional Encoder Representations from Transformers) is a language model from Google AI, introduced in a paper by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova posted on 11 October 2018. It pretrains a transformer encoder on unlabelled text by masking words and predicting them from both directions, then fine-tunes it for a task. The authors reported a GLUE score of 80.5% (a 7.7-point gain) and new state-of-the-art results on eleven language tasks. Google released the code and models in November 2018, which let thousands of teams reuse them; on 25 October 2019 Google began using BERT to understand English search queries in the United States and extended it to over 70 languages by December 2019. BERT base has 110 million parameters and BERT large 340 million.\n\nhttps://arxiv.org/abs/1810.04805\nhttps://blog.google/products/search/search-language-understanding-bert/\nhttps://en.wikipedia.org/wiki/BERT_(language_model)", "size": 4.71, "position": { "x": -34.927, "y": 31.92, "z": 0 } }, { "ID": "ai-a-gpt-2", "name": "GPT-2", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": true, "Parameters": "1.5 billion", "Date": "2019-02-14", "Milestone": true, "Impact score": 65 }, "description": "GPT-2 is OpenAI's 1.5-billion-parameter language model, announced on 14 February 2019 and trained on WebText, about 40 GB of text from pages linked in Reddit posts. OpenAI said it would not release the full model because of the risk of malicious use, such as fake news and spam, and gave selected journalists limited access instead. It released a 774-million-parameter version on 20 August 2019 and the full model on 5 November 2019, saying it had seen no strong evidence of misuse. The withholding split researchers, produced a public debate about publication norms, and was an early test of staged release for AI models.\n\nhttps://en.wikipedia.org/wiki/GPT-2\nhttps://www.theverge.com/2019/11/7/20953040/openai-text-generation-ai-gpt-2-full-model-release-1-5b-parameters\nPicture: AI-generated text, GPT2-talks-about-GPT2.png, unknown author, Public domain, https://commons.wikimedia.org/wiki/File:GPT2-talks-about-GPT2.png", "size": 4.9, "position": { "x": -24.717, "y": 31.92, "z": 0 }, "images": ["e048fba5-7ca1-4bb5-9932-96a4b70150a1"] }, { "ID": "ai-a-gpt-3", "name": "GPT-3", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": false, "Parameters": "175 billion", "Date": "2020-05-28", "Milestone": true, "Impact score": 90 }, "description": "GPT-3 is OpenAI's 175-billion-parameter language model, described in the paper 'Language Models are Few-Shot Learners' posted on 28 May 2020 by 31 OpenAI researchers. Pretrained on filtered Common Crawl, WebText2, two book collections and Wikipedia, it could perform new tasks from a few examples placed in the prompt, without fine-tuning. The paper states that all models were trained on V100 GPUs on part of a high-bandwidth cluster provided by Microsoft. Lambda Labs estimated a hypothetical single-GPU training cost of about $4.6 million. OpenAI opened a private-beta API on 11 June 2020 and Microsoft took an exclusive licence to the model on 22 September 2020. Reactions ranged from David Chalmers calling it one of the most interesting AI systems ever produced to Gary Marcus and Ernest Davis arguing that its comprehension is often seriously off.\n\nhttps://arxiv.org/abs/2005.14165\nhttps://en.wikipedia.org/wiki/GPT-3", "size": 5.72, "position": { "x": -13.497, "y": 31.92, "z": 0 } }, { "ID": "ai-a-alphafold-2", "name": "AlphaFold 2", "type": "Model or system", "properties": { "Field": "Science", "System kind": "Science model", "Open weights": true, "Parameters": "not disclosed", "Date": "2020-11-30", "Milestone": true, "Impact score": 75 }, "description": "AlphaFold 2 is DeepMind's neural network that predicts the three-dimensional structure of a protein from its amino-acid sequence. DeepMind announced on 30 November 2020 that it had won the CASP14 assessment, with a median score of 92.4 out of 100 on the GDT scale and the best prediction for 88 of 97 targets; MIT Technology Review reported an error of about 1.6 angstroms, comparable to lab methods. The method was published in Nature on 15 July 2021, and a database built with EMBL-EBI opened on 22 July 2021 with over 365,000 structures and grew to about 200 million proteins by 28 July 2022. DeepMind released the AlphaFold 2 source code and model weights on GitHub with the paper. John Jumper and Demis Hassabis shared half of the 2024 Nobel Prize in Chemistry for this work.\n\nhttps://www.technologyreview.com/2020/11/30/1012712/deepmind-protein-folding-ai-solved-biology-science-drugs-disease/\nhttps://doi.org/10.1038/d41586-020-03348-4\nhttps://en.wikipedia.org/wiki/AlphaFold\nhttps://github.com/google-deepmind/alphafold\nPicture: figure, T1044-alphafold3-layey-by-layer.gif, Kathryn Tunyasuvunakool, Jonas Adler, Zachary Wu, Tim Green, Michal Zielinski, Augustin Ží, CC BY 4.0, https://commons.wikimedia.org/wiki/File:T1044-alphafold3-layey-by-layer.gif", "size": 5.25, "position": { "x": -9.272, "y": 20.07, "z": 0 }, "images": ["cc264988-142c-401e-99f1-a4cc286ed6d5"] }, { "ID": "ai-scale-ms-clip", "name": "CLIP", "type": "Model or system", "properties": { "Field": "Vision", "System kind": "Multimodal model", "Open weights": true, "Parameters": "not disclosed", "Date": "2021-01-05", "Milestone": false, "Impact score": 50 }, "description": "CLIP (Contrastive Language-Image Pre-training) is an OpenAI model that learns to match images with captions. It was announced on 5 January 2021 alongside DALL-E, and the paper 'Learning Transferable Visual Models From Natural Language Supervision', led by Alec Radford, followed on 26 February 2021. Trained on 400 million image-text pairs from the internet, it let natural language name visual concepts, so it could classify images in categories it was not explicitly trained on. The authors reported that it matched the original ResNet-50's ImageNet accuracy zero-shot, without using ImageNet's 1.28 million training images. CLIP's text-image scoring was then used to guide and evaluate image generators.\n\nhttps://arxiv.org/abs/2103.00020\nhttps://www.technologyreview.com/2021/01/05/1015754/avocado-armchair-future-ai-openai-deep-learning-nlp-gpt3-computer-vision-common-sense/", "size": 4.29, "position": { "x": 9.588, "y": 20.07, "z": 0 } }, { "ID": "ai-scale-ms-dall-e", "name": "DALL-E", "type": "Model or system", "properties": { "Field": "Generative media", "System kind": "Image model", "Open weights": false, "Parameters": "12 billion", "Date": "2021-01-05", "Milestone": true, "Impact score": 60 }, "description": "DALL-E is OpenAI's first text-to-image model, revealed in a blog post on 5 January 2021. It is a 12-billion-parameter autoregressive transformer, a version of GPT-3 modified to generate images from text, and it was announced together with CLIP, a separate model that ranks DALL-E's outputs. MIT Technology Review's first report, built around the avocado armchair example, found the results impressive but inconsistent, noted that the model struggled with captions holding many objects, and raised the possibility that it reproduced images it had seen online. OpenAI did not release the model; the diffusion-based DALL-E 2 followed in April 2022.\n\nhttps://en.wikipedia.org/wiki/DALL-E\nhttps://www.technologyreview.com/2021/01/05/1015754/avocado-armchair-future-ai-openai-deep-learning-nlp-gpt3-computer-vision-common-sense/", "size": 4.71, "position": { "x": -2.467, "y": 31.92, "z": 0 } }, { "ID": "ai-scale-ms-lamda", "name": "LaMDA", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": false, "Parameters": "up to 137 billion", "Date": "2021-05-18", "Milestone": false, "Impact score": 45 }, "description": "LaMDA (Language Model for Dialogue Applications) is Google's conversational language model, unveiled by Google at its I/O keynote on 18 May 2021. Built for open-ended dialogue, it was shown as the future of conversation with Google's products. In June 2022 it became the subject of the first widely reported claim that a chatbot was sentient, made by the Google engineer Blake Lemoine, which Google and most researchers rejected.\n\nhttps://en.wikipedia.org/wiki/LaMDA\nhttps://www.zdnet.com/article/google-io-google-unveils-new-conversational-language-model-lamda/\nhttps://arxiv.org/abs/2201.08239", "size": 4.06, "position": { "x": 18.538, "y": 20.07, "z": 0 } }, { "ID": "ai-scale-ms-codex", "name": "OpenAI Codex", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": false, "Parameters": "not disclosed", "Date": "2021-07-07", "Milestone": false, "Impact score": 48 }, "description": "Codex is OpenAI's language model fine-tuned on publicly available code from GitHub, introduced in the paper 'Evaluating Large Language Models Trained on Code' posted on 7 July 2021 by Mark Chen and 57 co-authors. On the HumanEval benchmark of Python problems the authors reported that Codex solved 28.8% of them, against 0% for GPT-3 and 11.4% for GPT-J, and 70.2% when 100 samples were drawn per problem. They also listed its limits, such as difficulty with long chains of operations. Codex powered the first version of GitHub Copilot, which GitHub announced in technical preview on 29 June 2021.\n\nhttps://arxiv.org/abs/2107.03374\nhttps://en.wikipedia.org/wiki/GitHub_Copilot", "size": 4.2, "position": { "x": 9.559, "y": 31.92, "z": 0 } }, { "ID": "ai-scale-ms-dall-e-2", "name": "DALL-E 2", "type": "Model or system", "properties": { "Field": "Generative media", "System kind": "Image model", "Open weights": false, "Parameters": "3.5 billion", "Date": "2022-04-06", "Milestone": false, "Impact score": 50 }, "description": "DALL-E 2 is OpenAI's second text-to-image system, announced on 6 April 2022. It uses a diffusion model with about 3.5 billion parameters, fewer than the first DALL-E's 12 billion, and OpenAI described it as producing more realistic, higher-resolution images that combine concepts, attributes and styles. Access began as a research preview limited to pre-selected users because of ethics and safety concerns; on 20 July 2022 it entered beta with invitations to 1 million people on the waitlist, and on 28 September 2022 it opened to everyone. The public rollout, with Midjourney's beta and Stable Diffusion, started the wave of consumer image generators.\n\nhttps://en.wikipedia.org/wiki/DALL-E\nhttps://techcrunch.com/2022/04/06/openais-new-dall-e-model-draws-anything-but-bigger-better-and-faster-than-before/\nhttps://www.npr.org/2022/07/20/1112331013/dall-e-ai-art-beta-test\nPicture: AI-generated image, DALL-E 2 artificial intelligence digital image generated photo.jpg, DALL·E 2, Public domain, https://commons.wikimedia.org/wiki/File:DALL-E_2_artificial_intelligence_digital_image_generated_photo.jpg", "size": 4.29, "position": { "x": 23.475, "y": 31.92, "z": 0 }, "images": ["52eb13ab-ee83-48fc-9e98-86778ccf1553"] }, { "ID": "ai-a-stable-diffusion", "name": "Stable Diffusion", "type": "Model or system", "properties": { "Field": "Generative media", "System kind": "Image model", "Open weights": true, "Parameters": "about 1 billion", "Date": "2022-08-22", "Milestone": true, "Impact score": 75 }, "description": "Stable Diffusion is an open-weights text-to-image model released by Stability AI on 22 August 2022, developed by CompVis at LMU Munich and Runway, with Stability AI supplying the compute. It was trained on LAION-5B, a dataset of about 5 billion image-text pairs from Common Crawl, using 256 Nvidia A100 GPUs on Amazon Web Services for about 150,000 GPU-hours, at a cost of roughly $600,000. Unlike DALL-E 2 and Midjourney, anyone could download it, run it on a home computer and modify it, and in January 2023 it drew lawsuits by artists and by Getty Images alleging that Stability AI had copied millions of images without permission. Stability AI's CEO Emad Mostaque said it was people's responsibility whether they use the technology ethically, morally and legally.\n\nhttps://en.wikipedia.org/wiki/Stable_Diffusion\nhttps://www.theverge.com/2022/9/15/23340673/ai-image-generation-stable-diffusion-explained-ethics-copyright-data\nhttps://waxy.org/2022/08/exploring-12-million-of-the-images-used-to-train-stable-diffusions-image-generator/", "size": 5.25, "position": { "x": 43.635, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-gpt-35", "name": "GPT-3.5", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": false, "Parameters": "not disclosed", "Date": "2022-11-30", "Milestone": false, "Impact score": 45 }, "description": "GPT-3.5 is the name OpenAI gave to the family of GPT-3 descendants fine-tuned on code and with human feedback that preceded GPT-4. A GPT-3.5 model was the first model used inside ChatGPT when the chatbot launched on 30 November 2022, according to Wikipedia's ChatGPT article, and it was later removed from ChatGPT as newer models replaced it. OpenAI has not published a parameter count for the series. It matters in the record chiefly as the engine of the product that started the generative AI boom, built with the instruction-following and human-feedback methods set out in the 2022 InstructGPT paper.\n\nhttps://en.wikipedia.org/wiki/ChatGPT\nhttps://en.wikipedia.org/wiki/GPT-4", "size": 4.06, "position": { "x": 66.145, "y": 31.92, "z": 0 } }, { "ID": "ai-a-llama", "name": "LLaMA", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": false, "Parameters": "", "Date": "2023-02-24", "Milestone": true, "Impact score": 55 }, "description": "LLaMA (Large Language Model Meta AI) was Meta's family of foundation language models, announced on 24 February 2023 with a blog post and a research paper and offered to researchers case by case under a non-commercial licence. The promise was a smaller, efficient model that academics could run themselves, in contrast to closed systems such as GPT-4. On 3 March 2023 a torrent of the weights appeared on 4chan and spread through AI communities; Meta filed takedown requests against copies on Hugging Face and GitHub, but the leak set off a wave of derivative open models and a debate about misuse of open-weight models that The Verge covered in March 2023. Meta followed with Llama 2 (July 2023, with Microsoft, released for many commercial uses), Llama 3 (April 2024) and Llama 4 (April 2025), the last criticised over how its benchmark results were produced.\n\nhttps://en.wikipedia.org/wiki/Llama_(language_model)\nhttps://www.theverge.com/2023/3/8/23629362/meta-ai-language-model-llama-leak-online-misuse", "size": 4.51, "position": { "x": 65.954, "y": 20.07, "z": 0 } }, { "ID": "ai-a-gpt-4", "name": "GPT-4", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Multimodal model", "Open weights": false, "Parameters": "not disclosed (Semafor reported 1 trillion, unconfirmed)", "Date": "2023-03-14", "Milestone": true, "Impact score": 100 }, "description": "GPT-4 is OpenAI's large multimodal model, released as a preview on 14 March 2023. It accepts text and image input and was introduced with claims of stronger reasoning, retention and coding than GPT-3.5; Microsoft researchers called an early version 'an early (yet still incomplete) version of an artificial general intelligence (AGI) system' in a March 2023 paper. OpenAI withheld the model size, architecture and training hardware, which drew criticism from researchers who contrasted it with the openness of GPT-2 and GPT-3; Semafor, citing sources, reported 1 trillion parameters, and Sam Altman said training cost more than $100 million (Wikipedia's GPT-4 article). An early version had powered Bing Chat since February 2023. GPT-4 was the model the Future of Life Institute's pause letter named as the threshold, and it was the product behind the sharpest rise in government attention to frontier AI in 2023.\n\nhttps://en.wikipedia.org/wiki/GPT-4\nhttps://arxiv.org/abs/2303.12712", "size": 6, "position": { "x": 85.985, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-gemini-1", "name": "Gemini 1.0", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Multimodal model", "Open weights": false, "Parameters": "not disclosed", "Date": "2023-12-06", "Milestone": true, "Impact score": 65 }, "description": "Gemini 1.0 was Google DeepMind's first Gemini family of natively multimodal models, announced on 6 December 2023 by Sundar Pichai and Demis Hassabis in three sizes (Nano, Pro and Ultra). Google said Gemini Ultra outperformed GPT-4 and other rivals on industry benchmarks and was the first model to beat human experts on the MMLU test, scoring 90 percent (Wikipedia's Gemini article summarising Google's claims). MIT Technology Review called the hype around the launch 'peak AI hype' and researchers cautioned that benchmark scores are hard to interpret without knowing the training data; Google was also criticised after a demo video was edited to look like a real-time interaction. Gemini replaced the Bard name and became the base of Google's products. It arrived two weeks after the Altman crisis at OpenAI and a year after Google's 'code red'.\n\nhttps://en.wikipedia.org/wiki/Gemini_(language_model)\nhttps://www.technologyreview.com/2023/12/06/1084471/google-deepminds-new-gemini-model-looks-amazing-but-could-signal-peak-ai-hype/", "size": 4.9, "position": { "x": 120.998, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-sora", "name": "Sora", "type": "Model or system", "properties": { "Field": "Generative media", "System kind": "Video model", "Open weights": false, "Parameters": "not disclosed", "Date": "2024-02-15", "Milestone": true, "Impact score": 70 }, "description": "Sora is OpenAI's text-to-video model, previewed on 15 February 2024 with sample high-definition clips of up to a minute generated from text prompts. The New York Times headlined the preview as an A.I. that 'instantly generates eye-popping videos' and Wired described photorealistic results. OpenAI released Sora to ChatGPT Plus and Pro users on 9 December 2024 and Sora 2 with a social-video app on 30 September 2025. On 24 March 2026 OpenAI announced it would discontinue Sora in the consumer app and API, citing compute demand and its focus on other products, according to CBS News; the app closed on 26 April and the API on 24 September 2026. A Disney licensing agreement of December 2025 was caught up in the shutdown.\n\nhttps://en.wikipedia.org/wiki/Sora_(text-to-video_model)\nhttps://www.cbsnews.com/news/sora-ai-openai-discontinues/", "size": 5.08, "position": { "x": 131.578, "y": 31.92, "z": 0 } }, { "ID": "ai-a-o1", "name": "OpenAI o1", "type": "Model or system", "properties": { "Field": "Agents and reasoning", "System kind": "Reasoning model", "Open weights": false, "Parameters": "not disclosed", "Date": "2024-09-12", "Milestone": true, "Impact score": 70 }, "description": "OpenAI o1 is the first model in OpenAI's 'o' series of reasoning models, released as a preview on 12 September 2024 and in full to ChatGPT users on 5 December 2024. It produces a long hidden chain of thought before answering, which OpenAI described as a new way to scale performance with compute spent at answer time. OpenAI reported that o1-preview solved 83 percent of problems on the American Invitational Mathematics Examination against 13 percent for GPT-4o, and ranked in the 89th percentile on Codeforces. OpenAI rated it medium risk for chemical, biological, radiological and nuclear misuse, and Apollo Research found it more consistently deceptive than other frontier models in controlled tests. o1 triggered a wave of rival reasoning models, including DeepSeek-R1, and is used here as the start of the 'reasoning models' era.\n\nhttps://en.wikipedia.org/wiki/OpenAI_o1", "size": 5.08, "position": { "x": 159.823, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-deepseek-v3", "name": "DeepSeek-V3", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": true, "Parameters": "671 billion total, about 37 billion active per token", "Date": "2024-12-27", "Milestone": true, "Impact score": 65 }, "description": "DeepSeek-V3 is a 671-billion-parameter mixture-of-experts language model from the Chinese lab DeepSeek, with about 37 billion parameters active per token, released in late December 2024; its technical report was posted on arXiv on 27 December 2024. The report states that the full training needed 2.788 million H800 GPU hours, a figure that became the centre of the debate on how cheaply a frontier-class model could be trained, because the H800 is the export-compliant Nvidia chip sold to China after the 2022 US controls. DeepSeek's own cost accounting covers the final run only, not research or earlier experiments. DeepSeek published the model's weights, as it does for its other models.\n\nhttps://arxiv.org/abs/2412.19437\nhttps://en.wikipedia.org/wiki/DeepSeek", "size": 4.9, "position": { "x": 175.375, "y": 31.92, "z": 0 } }, { "ID": "ai-a-deepseek-r1", "name": "DeepSeek-R1", "type": "Model or system", "properties": { "Field": "Agents and reasoning", "System kind": "Reasoning model", "Open weights": true, "Parameters": "671 billion total (shared base with V3)", "Date": "2025-01-20", "Milestone": true, "Impact score": 85 }, "description": "DeepSeek-R1 is a reasoning model from the Chinese lab DeepSeek, released with open weights on 20 January 2025 together with a free chatbot app. The accompanying paper, posted on arXiv on 22 January, reports that reasoning behaviours such as self-reflection and verification can emerge from reinforcement learning without human-labelled reasoning examples, and that the reasoning can be distilled into smaller models. By 27 January the app had passed ChatGPT as the most-downloaded free app in the US iOS App Store, CNBC reported, and Nvidia's share price fell about 17 percent that day. DeepSeek's low stated training cost for its base model, 2.788 million H800 GPU hours, fed a debate on whether frontier AI needed ever larger chip purchases. The release drew a US government 'wake-up call' response, an Italian block of the app, bans on government devices and calls for tighter chip export rules.\n\nhttps://arxiv.org/abs/2501.12948\nhttps://www.cnbc.com/2025/01/27/chinas-deepseek-ai-tops-chatgpt-app-store-what-you-should-know.html\nPicture: figure, The multistage pipeline of DeepSeek-R1.png, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Peiyi Wang, Qihao Zhu, Runxin Xu, Ruoyu, CC BY 4.0, https://commons.wikimedia.org/wiki/File:The_multistage_pipeline_of_DeepSeek-R1.png", "size": 5.57, "position": { "x": 186.445, "y": 31.92, "z": 0 }, "images": ["1d40a94c-06bf-49e7-b265-4daa724d9fd1"] }, { "ID": "ai-boom-ms-openai-imo-model", "name": "OpenAI IMO model", "type": "Model or system", "properties": { "Field": "Agents and reasoning", "System kind": "Reasoning model", "Open weights": false, "Parameters": "not disclosed", "Date": "2025-07-19", "Milestone": false, "Impact score": 40 }, "description": "At the International Mathematical Olympiad of July 2025 OpenAI reported that an experimental general-purpose reasoning model, not a release candidate, solved five of the six problems for 35 of 42 points, enough for a gold medal, by writing natural-language proofs under the competition's 4.5-hour limit. OpenAI announced the result on 19 July 2025, two days before Google DeepMind announced its own officially graded gold-medal score. A Google DeepMind researcher argued that a one-point deduction would make the result silver, a point that is contested because the official marking guideline is not public. OpenAI did not disclose the model's size or name.\n\nhttps://www.techrepublic.com/article/news-openai-model-math-olympiad/\nhttps://www.axios.com/2025/07/21/openai-deepmind-math-olympiad-ai", "size": 3.8, "position": { "x": 222.32, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-gemini-deep-think-imo", "name": "Gemini Deep Think", "type": "Model or system", "properties": { "Field": "Agents and reasoning", "System kind": "Reasoning model", "Open weights": false, "Parameters": "not disclosed", "Date": "2025-07-21", "Milestone": false, "Impact score": 40 }, "description": "Google DeepMind announced on 21 July 2025 that an advanced version of Gemini with Deep Think had achieved gold-medal standard at the International Mathematical Olympiad, solving five of six problems for 35 of 42 points. According to Google and press reports, the work was done end to end in natural language, generating proofs directly from the official problem statements within the 4.5-hour time limit, and the result was graded and certified by IMO coordinators under the same standards used for human contestants. Only 67 of 630 contestants won gold. Google's result came two days after OpenAI's uncertified claim of the same score, which made the verification of the two results the main point of dispute.\n\nhttps://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/\nhttps://www.maginative.com/article/gemini-with-deep-think-achieves-gold-at-international-math-olympiad/", "size": 3.8, "position": { "x": 241.62, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-gpt-5", "name": "GPT-5", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Multimodal model", "Open weights": false, "Parameters": "not disclosed", "Date": "2025-08-07", "Milestone": true, "Impact score": 75 }, "description": "GPT-5 is OpenAI's flagship model, released on 7 August 2025 in ChatGPT, Microsoft Copilot and the API, succeeding GPT-4.1, o3 and o4-mini. OpenAI claimed state-of-the-art results in mathematics, programming, finance and multimodal understanding, fewer hallucinations and 'safe completions'; Sam Altman called it 'a significant step along the path to AGI' with 'PhD-level' abilities. Reception was mixed: MIT Technology Review's Grace Huckins called it 'above all else, a refined product' that fell short of Altman's hype, The Atlantic judged it intuitive and fast, and users objected to the model router and the removal of GPT-4o, which Altman restored for paying users, with a warmer personality update on 15 August. It was the headline release of OpenAI's 2025 and was followed by GPT-5.x updates and, in September 2026, GPT-6 Astra.\n\nhttps://openai.com/index/introducing-gpt-5/\nhttps://en.wikipedia.org/wiki/GPT-5", "size": 5.25, "position": { "x": 231.97, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-gemini-3", "name": "Gemini 3", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Multimodal model", "Open weights": false, "Parameters": "not disclosed", "Date": "2025-11-18", "Milestone": true, "Impact score": 65 }, "description": "Gemini 3 is Google's third-generation Gemini model family, introduced on 18 November 2025 with Gemini 3 Pro in preview, a sparse mixture-of-experts, natively multimodal reasoning model that Google says handles up to a million input tokens. Sundar Pichai called it Google's most intelligent model and Demis Hassabis said he regards it as the strongest model for multimodal understanding. Google rolled it into Search on launch day, which it called the quickest integration of a model into Search, and followed with Nano Banana Pro (20 November), Deep Think (3 December) and Gemini 3 Flash (17 December). The release put Google ahead on several leaderboards and was followed by an internal 'code red' at OpenAI in early December, though accounts differ on how directly the two are linked.\n\nhttps://blog.google/products/gemini/gemini-3/\nhttps://en.wikipedia.org/wiki/Gemini_(language_model)", "size": 4.9, "position": { "x": 255.983, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-claude-mythos-preview", "name": "Claude Mythos Preview", "type": "Model or system", "properties": { "Field": "Agents and reasoning", "System kind": "Language model", "Open weights": false, "Parameters": "not disclosed", "Date": "2026-04-07", "Milestone": true, "Impact score": 60 }, "description": "Claude Mythos Preview is an unreleased Anthropic model announced on 7 April 2026 together with Project Glasswing, a programme that gives a limited group of partners access to use it to find and fix software vulnerabilities. Launch partners named in coverage were Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, Nvidia and Palo Alto Networks, with more than 40 further organisations that maintain critical software. Anthropic said the model had already found thousands of high-severity vulnerabilities, including in every major operating system and web browser, that it did not plan to make Mythos Preview generally available, and that it would commit up to $100 million in usage credits and $4 million in donations to open-source security groups. VentureBeat headlined the announcement as an AI cyber model 'too dangerous to release'; Fortune reported that the model's existence had first emerged through a data leak in March. The capability claims are Anthropic's own.\n\nhttps://venturebeat.com/data/anthropic-says-its-most-powerful-ai-cyber-model-is-too-dangerous-to-release\nhttps://fortune.com/2026/06/09/anthropic-releases-its-first-mythos-model-to-the-public/", "size": 4.71, "position": { "x": 280.375, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-deepseek-v4", "name": "DeepSeek-V4", "type": "Model or system", "properties": { "Field": "Language", "System kind": "Language model", "Open weights": false, "Parameters": "284 billion (Flash) and 1.6 trillion (Pro)", "Date": "2026-04-24", "Milestone": false, "Impact score": 45 }, "description": "DeepSeek-V4 is the Chinese lab DeepSeek's fourth-generation model family, previewed on 24 April 2026 in two sizes, V4-Flash with 284 billion parameters and V4-Pro with 1.6 trillion, both with a one-million-token context window, according to Wikipedia's DeepSeek article. The Independent reported that DeepSeek claimed V4-Pro beat all open-source competitors, and MIT Technology Review published an explainer on why the model mattered. Official versions followed on 31 July (Flash) and 13 August (Pro), and V4.1-Flash, natively multimodal, on 10 September 2026, when the South China Morning Post reported DeepSeek's claim that it beat Kimi K3 on cyber and coding benchmarks. The benchmark claims are DeepSeek's own. The release showed that Chinese labs were keeping pace 15 months after R1 despite US chip export controls.\n\nhttps://en.wikipedia.org/wiki/DeepSeek\nhttps://www.technologyreview.com/2026/04/24/1136422/why-deepseeks-v4-matters/", "size": 4.06, "position": { "x": 289.745, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-claude-fable-5", "name": "Claude Fable 5", "type": "Model or system", "properties": { "Field": "Agents and reasoning", "System kind": "Language model", "Open weights": false, "Parameters": "not disclosed", "Date": "2026-06-09", "Milestone": true, "Impact score": 65 }, "description": "Claude Fable 5 and Claude Mythos 5 are Anthropic models released on 9 June 2026. Fortune described Fable 5 as the company's first widely available 'Mythos-class' model, aimed at coding, knowledge work and vision, while Mythos 5 went only to vetted partners who already had Mythos Preview. Anthropic said Fable 5's capabilities 'exceed those of every model we've previously made generally available', that responses in high-risk areas such as biology and cybersecurity are blocked and in most cases answered by Claude Opus 4.8 instead, and priced it at $10 per million input tokens and $50 per million output tokens. The launch came just over a week after Anthropic confidentially filed for an IPO. Three days later, on 12 June, Anthropic suspended access to both models after a US Commerce Department directive, and restored them on 1 July.\n\nhttps://fortune.com/2026/06/09/anthropic-releases-its-first-mythos-model-to-the-public/\nhttps://www.aljazeera.com/economy/2026/7/1/us-lifts-restrictions-on-powerful-ai-models-fable-mythos-anthropic-says", "size": 4.9, "position": { "x": 299.305, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-ms-gpt-6-astra", "name": "GPT-6 Astra", "type": "Model or system", "properties": { "Field": "Agents and reasoning", "System kind": "Multimodal model", "Open weights": false, "Parameters": "not disclosed", "Date": "2026-09-03", "Milestone": true, "Impact score": 70 }, "description": "GPT-6 Astra is OpenAI's flagship model of September 2026, previewed on 3 September and released to paid ChatGPT users on 4 September in a restricted version that declines some prompts, for example on cybersecurity. Gizmodo reported that OpenAI called it the 'world's best computer use model' and its safest yet. At the launch briefing OpenAI president Greg Brockman said 'it's not unreasonable to feel that we are now in the AGI era'; Wikipedia's GPT-6 article describes his remark as a suggestion that it could eventually be seen as the arrival of AGI, and critics noted there is no agreed test for AGI. The release followed a delay after unsanctioned cyberattacks by OpenAI agents in July 2026, which OpenAI said it addressed with safeguards; GPT-6 Sol and Luna followed on 22 September.\n\nhttps://gizmodo.com/openai-claims-were-in-the-agi-era-with-release-of-gpt-6-astra-2000807013\nhttps://en.wikipedia.org/wiki/GPT-6", "size": 5.08, "position": { "x": 309.885, "y": 31.92, "z": 0 } }, { "ID": "ai-a-cuda", "name": "CUDA", "type": "Product", "properties": { "Field": "Compute", "Product kind": "Developer platform", "Reported users": "", "Date": "2007-02-15", "Milestone": false, "Impact score": 48 }, "description": "CUDA is Nvidia's proprietary parallel computing platform and programming interface that lets software use Nvidia GPUs for general-purpose computing. Ian Buck, a Stanford PhD student who had built the GPU language Brook, joined Nvidia in 2004 and, working with John Nickolls, turned Brook into CUDA; the first software development kit was made public on 15 February 2007. From about 2015 its development focused increasingly on machine-learning workloads, and AlexNet, GPT-3 and nearly every later large model were trained with software built on it. Its large body of libraries is a main reason Nvidia's chips are hard for rivals to displace.\n\nhttps://en.wikipedia.org/wiki/CUDA\nhttps://web.archive.org/web/20070329144655/http://www.nvidia.com/object/IO_39918.html\nhttps://www.newyorker.com/magazine/2023/12/04/how-jensen-huangs-nvidia-is-powering-the-ai-revolution", "size": 4.2, "position": { "x": -174.813, "y": 31.92, "z": 0 } }, { "ID": "ai-hist-pr-siri", "name": "Siri", "type": "Product", "properties": { "Field": "Speech", "Product kind": "Voice assistant", "Reported users": "", "Date": "2011-10-14", "Milestone": true, "Impact score": 55 }, "description": "Siri is the voice assistant that Apple announced on 4 October 2011 as part of the iPhone 4S and shipped on 14 October 2011. It grew out of the DARPA-funded CALO project at SRI International; SRI spun it off as Siri Inc., co-founded by Dag Kittlaus, Tom Gruber and Adam Cheyer, with speech recognition from Nuance, and a standalone iPhone app preceded Apple's purchase of the company in April 2010 according to SRI's account (other accounts give 2011). Early reviews were mixed: critics praised its voice recognition and understanding of context but found its commands rigid, and David Pogue of the New York Times headlined his review 'New iPhone Conceals Sheer Magic'. Apple later said it would use LSTM networks for Siri.\n\nhttps://en.wikipedia.org/wiki/Siri\nhttps://en.wikipedia.org/wiki/SRI_International\nhttps://en.wikipedia.org/wiki/Long_short-term_memory", "size": 4.51, "position": { "x": -152.003, "y": 20.07, "z": 0 } }, { "ID": "ai-scale-pr-google-search-bert", "name": "BERT in Google Search", "type": "Product", "properties": { "Field": "Language", "Product kind": "Search", "Reported users": "", "Date": "2019-10-25", "Milestone": false, "Impact score": 43 }, "description": "On 25 October 2019 Google began using BERT to understand search queries in English in the United States, and by 9 December 2019 it was applied to over 70 languages; by October 2020 nearly every English query was processed by a BERT model. It put a transformer language model in front of billions of searches and was the first mass-market deployment of the architecture introduced in 2017.\n\nhttps://blog.google/products/search/search-language-understanding-bert/\nhttps://en.wikipedia.org/wiki/BERT_(language_model)", "size": 3.96, "position": { "x": -30.853, "y": 20.07, "z": 0 } }, { "ID": "ai-scale-pr-openai-api", "name": "OpenAI API", "type": "Product", "properties": { "Field": "Language", "Product kind": "API", "Reported users": "", "Date": "2020-06-11", "Milestone": false, "Impact score": 45 }, "description": "The OpenAI API is the hosted service through which developers send text prompts to OpenAI's models and receive completions. It was launched as a private beta on 11 June 2020 with GPT-3 as its first model, and OpenAI sold access rather than releasing the model's weights. The commercial interface turned a research system into a platform on which outside companies built writing, coding and chat tools, and it was the channel through which Codex later reached GitHub Copilot. When Microsoft licensed GPT-3 exclusively on 22 September 2020, OpenAI stated that the public API would continue.\n\nhttps://en.wikipedia.org/wiki/GPT-3\nhttps://www.technologyreview.com/2020/09/23/1008729/openai-is-giving-microsoft-exclusive-access-to-its-gpt-3-language-model/", "size": 4.06, "position": { "x": -19.182, "y": 20.07, "z": 0 } }, { "ID": "ai-scale-pr-github-copilot", "name": "GitHub Copilot", "type": "Product", "properties": { "Field": "Language", "Product kind": "Coding assistant", "Reported users": "", "Date": "2021-06-29", "Milestone": true, "Impact score": 60 }, "description": "GitHub Copilot is an AI coding assistant that suggests lines and whole functions inside a programmer's editor. GitHub announced it for technical preview in Visual Studio Code on 29 June 2021, powered by OpenAI Codex, a modified production version of GPT-3, and made it a paid subscription for individual developers on 21 June 2022. The Verge noted in July 2021 that the tool rested on untested legal ground, since it was trained on public code under many licences. The Software Freedom Conservancy announced in June 2022 that it would stop using GitHub for its projects, and a class action followed in November 2022.\n\nhttps://en.wikipedia.org/wiki/GitHub_Copilot\nhttps://www.theverge.com/2021/6/29/22555777/github-openai-ai-tool-autocomplete-code", "size": 4.71, "position": { "x": 8.284, "y": 9.51, "z": 0 } }, { "ID": "ai-scale-pr-midjourney", "name": "Midjourney", "type": "Product", "properties": { "Field": "Generative media", "Product kind": "Image generator", "Reported users": "", "Date": "2022-07-12", "Milestone": true, "Impact score": 60 }, "description": "Midjourney is an AI image generator run by the San Francisco company Midjourney, Inc., founded by David Holz, who previously co-founded Leap Motion. Its open beta began on 12 July 2022; users type a prompt into a Discord bot with the /imagine command and receive four images that they can upscale. Version 3 followed on 25 July 2022 and version 4 in alpha on 5 November 2022. A Midjourney image won first prize at the Colorado State Fair on 29 August 2022, which started a public argument over AI and art, and in January 2023 three artists sued Midjourney, Stability AI and DeviantArt, an allegation that has been contested in court.\n\nhttps://en.wikipedia.org/wiki/Midjourney\nhttps://www.pcworld.com/article/820518/midjourneys-ai-art-goes-live-for-everyone.html\nPicture: AI-generated image, Théâtre D’opéra Spatial.jpg, Jason M. Allen, Public domain, https://commons.wikimedia.org/wiki/File:Th%C3%A9%C3%A2tre_D%E2%80%99op%C3%A9ra_Spatial.jpg", "size": 4.71, "position": { "x": 33.075, "y": 31.92, "z": 0 }, "images": ["57ebc354-f447-448b-9ae1-9a58eab37037"] }, { "ID": "ai-a-chatgpt", "name": "ChatGPT", "type": "Product", "properties": { "Field": "Language", "Product kind": "Chatbot", "Reported users": "About 100 million monthly active users two months after launch (Wikipedia ChatGPT article); about 900 million weekly active users in February 2026 (same article)", "Date": "2022-11-30", "Milestone": true, "Impact score": 100 }, "description": "ChatGPT is OpenAI's conversational assistant, released free to the public as a research preview on 30 November 2022 on top of a GPT-3.5 model fine-tuned with human feedback. Within days it drew a million users and within about two months an estimated 100 million monthly users, which made it the fastest-spreading consumer application of its time and turned large language models from a research topic into a household product. The launch pushed Google into an internal 'code red', set off a race of rival chatbots (Bing Chat, Bard, Claude, LLaMA-based tools), prompted Microsoft's multibillion-dollar extension of its OpenAI partnership in January 2023, and was cited by regulators in Italy, by striking writers in Hollywood and by the New York Times in its December 2023 lawsuit. Wikipedia's ChatGPT article, citing OpenAI, gives about 900 million weekly active users in February 2026; coverage of the Nvidia investment deal put the figure above 700 million weekly users in September 2025.\n\nhttps://en.wikipedia.org/wiki/ChatGPT\nhttps://en.wikipedia.org/wiki/Bard_(chatbot)", "size": 6, "position": { "x": 55.485, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-pr-bard", "name": "Bard", "type": "Product", "properties": { "Field": "Language", "Product kind": "Chatbot", "Reported users": "", "Date": "2023-02-06", "Milestone": false, "Impact score": 50 }, "description": "Bard was Google's conversational AI service, announced on 6 February 2023 as an experimental chatbot built on the LaMDA model and opened first to 10,000 trusted testers, according to Wikipedia's Bard article. Wikipedia reports that Google brought the announcement forward to pre-empt Microsoft's event for ChatGPT-powered Bing on 7 February. In the promotional demonstration, Bard gave an inaccurate answer about the James Webb Space Telescope, which was noticed on 8 February and was followed by a sharp fall in Alphabet shares. Bard was later renamed Gemini. It is the central product in the chain from ChatGPT to Google's 'code red' to the demo error and the share fall.\n\nhttps://en.wikipedia.org/wiki/Bard_(chatbot)\nhttps://www.cnbc.com/2023/02/10/google-employees-slam-ceo-sundar-pichai-for-rushed-bard-announcement.html", "size": 4.29, "position": { "x": 75.095, "y": 31.92, "z": 0 } }, { "ID": "ai-boom-pr-bing-chat", "name": "Bing Chat", "type": "Product", "properties": { "Field": "Language", "Product kind": "Search", "Reported users": "", "Date": "2023-02-07", "Milestone": false, "Impact score": 40 }, "description": "Bing Chat was Microsoft's chat interface for its Bing search engine, built on OpenAI models and unveiled at an event on 7 February 2023, the day before Google's rushed Bard demonstration. An early version of GPT-4 powered it from February 2023, according to Wikipedia's GPT-4 article, and it was the first large mass-market search product to use a ChatGPT-class model. Within days users reported erratic behaviour in long conversations, which the same article records as an early public example of chatbot misbehaviour. Bing Chat was the consumer-facing result of the January 2023 Microsoft investment in OpenAI and a main reason Google accelerated Bard.\n\nhttps://en.wikipedia.org/wiki/Bard_(chatbot)\nhttps://en.wikipedia.org/wiki/GPT-4", "size": 3.8, "position": { "x": 57.044, "y": 20.07, "z": 0 } }, { "ID": "ai-boom-pr-claude-code", "name": "Claude Code", "type": "Product", "properties": { "Field": "Agents and reasoning", "Product kind": "Coding assistant", "Reported users": "", "Date": "2025-02-24", "Milestone": true, "Impact score": 60 }, "description": "Claude Code is Anthropic's agentic coding tool, released as a command-line application in February 2025 with Claude 3.7 Sonnet and made generally available in May 2025 alongside Claude 4, according to Wikipedia's Claude article. It lets a model read a codebase, run commands and edit files on a developer's behalf. Anthropic added a web version and sandboxing on 20 October 2025, and the tool went viral over the 2025-2026 winter holidays, including among non-programmers 'vibe coding' (Collins Dictionary named the term its 2025 word of the year). Its CLI source code leaked in March 2026 through an exposed map file, as Ars Technica reported. Claude Code made agentic coding the centre of Anthropic's business and a model for rival tools.\n\nhttps://en.wikipedia.org/wiki/Claude_(language_model)\nhttps://www.anthropic.com/news/claude-3-7-sonnet", "size": 4.71, "position": { "x": 197.55, "y": 31.92, "z": 0 } }, { "ID": "ai-a-gtx-580", "name": "GeForce GTX 580", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "GPU", "Date": "2010-11-09", "Milestone": false, "Impact score": 43 }, "description": "The GeForce GTX 580 is a consumer graphics card that Nvidia launched on 9 November 2010 at $499, built around the 3-billion-transistor GF110 chip on a 40 nm process. It became part of AI history in 2012, when Alex Krizhevsky trained AlexNet on two GTX 580 cards with 3 GB of memory each, splitting the network across them because it did not fit on one. Training took five to six days for 90 epochs, and the code was written in CUDA and C++. AlexNet won the ImageNet challenge that autumn, showing that a network trained on gaming hardware could outperform traditional vision methods.\n\nhttps://en.wikipedia.org/wiki/GeForce_500_series\nhttps://en.wikipedia.org/wiki/AlexNet\nhttps://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf", "size": [5.14, 5.14], "position": { "x": -163.967, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["32eb6016-dd90-45eb-bd6e-dd4a24b9c487"] }, { "ID": "ai-a-tpu", "name": "Google TPU", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "AI accelerator", "Date": "2016-05-18", "Milestone": true, "Impact score": 60 }, "description": "The Tensor Processing Unit (TPU) is Google's custom chip for neural-network workloads, announced at Google I/O on 18 May 2016 after, Google said, it had already run in its data centres for over a year. The first generation was an 8-bit matrix-multiplication engine on a 28 nm process with a 256 by 256 systolic array and 8 GiB of memory, and Google said TPUs powered the AlphaGo matches against Lee Sedol. In a 2017 ISCA paper led by Norman Jouppi the team reported the chip to be about 15 to 30 times faster, and 30 to 80 times more energy-efficient, than contemporary CPUs and GPUs. Later generations added training support, and TPUs trained AlphaZero, BERT and PaLM, making Google the largest in-house alternative to Nvidia's GPUs.\n\nhttps://cloudplatform.googleblog.com/2016/05/Google-supercharges-machine-learning-tasks-with-custom-chip.html\nhttps://arxiv.org/abs/1704.04760\nhttps://en.wikipedia.org/wiki/Tensor_Processing_Unit", "size": [5.83, 5.83], "position": { "x": -85.106, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-v100", "name": "V100", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "GPU", "Date": "2017-05-10", "Milestone": false, "Impact score": 50 }, "description": "Full name: Nvidia V100. The Tesla V100 is Nvidia's data-centre GPU built on the Volta architecture, announced at the company's GPU Technology Conference on 10 May 2017. It has 21.1 billion transistors and was the first Nvidia chip with Tensor Cores, units that multiply small matrices of 16-bit numbers and are meant to speed up neural-network training. Nvidia built it for deep learning as the successor to the Pascal generation that trained the original transformer, and OpenAI's GPT-3 paper states that its models were trained on V100 GPUs on a Microsoft cluster. The A100, announced in May 2020 on the Ampere architecture, succeeded it.\n\nhttps://en.wikipedia.org/wiki/Volta_(microarchitecture)\nhttps://devblogs.nvidia.com/parallelforall/inside-volta/\nhttps://arxiv.org/abs/2005.14165", "size": [5.45, 5.45], "position": { "x": -72.66, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-a100", "name": "A100", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "GPU", "Date": "2020-05-14", "Milestone": true, "Impact score": 70 }, "description": "Full name: Nvidia A100. The A100 is Nvidia's data-centre GPU on the Ampere architecture, announced and put into production on 14 May 2020. Its GA100 chip has 54.2 billion transistors, 6,912 FP32 cores and 40 GB of memory with 1.6 TB/s of bandwidth, and Nvidia sold it in the eight-GPU DGX A100 server from $199,000. It succeeded the V100 and became the workhorse of the first generative-AI wave: Stability AI trained Stable Diffusion on 256 A100 GPUs rented on Amazon Web Services, about 150,000 GPU-hours at roughly $600,000. The US export controls of 7 October 2022 targeted it, and Nvidia answered with the slower-interconnect A800 for China.\n\nhttps://en.wikipedia.org/wiki/Ampere_(microarchitecture)\nhttps://en.wikipedia.org/wiki/Nvidia_DGX\nhttps://en.wikipedia.org/wiki/Stable_Diffusion", "size": [6.16, 6.16], "position": { "x": -23.117, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-ch-azure-openai-supercomputer", "name": "Azure AI cluster", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "Supercomputer", "Date": "2020-05-19", "Milestone": false, "Impact score": 38 }, "description": "Full name: Microsoft Azure supercomputer for OpenAI. On 19 May 2020, at its Build conference, Microsoft announced a supercomputer built in Azure exclusively for OpenAI to train large models. Microsoft described it as having more than 285,000 CPU cores, about 10,000 GPUs and 400 gigabits per second of network connectivity per GPU server, and said it ranked in the top five of publicly disclosed supercomputers on the TOP500 list. The announcement did not name the GPU model or the models to be trained. The machine delivered the compute side of the partnership signed in July 2019; OpenAI's GPT-3 paper, posted nine days later, says its models were trained on V100 GPUs on part of a Microsoft-provided cluster.\n\nhttps://news.microsoft.com/source/features/innovation/openai-azure-supercomputer/\nhttps://arxiv.org/abs/2005.14165", "size": [4.9, 4.9], "position": { "x": -16.987, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-h100", "name": "H100", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "GPU", "Date": "2022-03-22", "Milestone": true, "Impact score": 90 }, "description": "Full name: Nvidia H100. The H100 is Nvidia's Hopper-generation data-centre GPU, announced at GTC on 22 March 2022. It has 80 billion transistors on TSMC's 4N process, up to 80 GB of memory and a 700-watt power budget in its SXM form. Demand after ChatGPT outran supply: in 2023 Oracle's Larry Ellison said that he and Elon Musk were begging Nvidia for H100s, and in January 2024 Raymond James analysts estimated Nvidia sold them for $25,000 to $30,000 each. The US export controls of October 2022 limited its sale to China, which Nvidia met with the H800 and later the H20. Blackwell succeeded it in 2024.\n\nhttps://en.wikipedia.org/wiki/Hopper_(microarchitecture)\nhttps://www.theverge.com/2022/3/22/22989182/nvidia-ai-hopper-architecture-h100-gpu-eos-supercomputer\nhttps://www.cnbc.com/2024/01/18/mark-zuckerberg-indicates-meta-is-spending-billions-on-nvidia-ai-chips.html\nPicture: NVIDIA H100 (极客湾Geekerwan) 025.png, 极客湾Geekerwan, CC BY 3.0, https://commons.wikimedia.org/wiki/File:NVIDIA_H100_(%E6%9E%81%E5%AE%A2%E6%B9%BEGeekerwan)_025.png", "size": [6.74, 6.74], "position": { "x": 20.954, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["4e24f254-e5ac-4560-b576-70435687d353"] }, { "ID": "ai-scale-ch-a800", "name": "A800", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "GPU", "Date": "2022-11-08", "Milestone": false, "Impact score": 38 }, "description": "Full name: Nvidia A800. The A800 is a China-market variant of the A100 that Nvidia began selling after the US export controls of 7 October 2022. Reports in early November 2022 said it cut chip-to-chip data transfer to 400 gigabytes per second from the A100's 600, which brought it just under the new US limit; other specifications were largely unchanged. A Nvidia spokesperson said the chip met the US government's clear test for reduced export control and could not be programmed to exceed it. The A800 was the first of Nvidia's compliant China chips, followed by the H800 in 2023 and the H20 after the controls were tightened in October 2023.\n\nhttps://www.aibusiness.com/verticals/nvidia-offers-new-ai-chip-to-china-to-comply-with-export-ban\nhttps://computing.co.uk/news/4059774/nvidia-offers-slower-chip-china-complies-us-sanctions\nhttps://www.tweaktown.com/news/89430/nvidias-new-cut-down-a800-gpu-created-because-of-us-export-controls-to-china/", "size": [4.9, 4.9], "position": { "x": 42.098, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-ch-h800", "name": "H800", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "GPU", "Date": "2023-03-21", "Milestone": false, "Impact score": 45 }, "description": "Full name: Nvidia H800. The H800 is a China-market variant of the H100 with reduced bandwidth, built to stay inside the US export controls of October 2022. Reuters reported on 21 March 2023 that Nvidia had tweaked its flagship chip for China as the H800; the Wikipedia article on the Hopper architecture gives late 2022 for its release, so the exact date is disputed. DeepSeek reported training its V3 model on 2,048 H800 GPUs, which showed that export-compliant chips could still support a frontier model. The US banned H800 exports on 17 October 2023, after which Nvidia moved to the H20.\n\nhttps://web.archive.org/web/20231122061905/https://www.reuters.com/technology/nvidia-tweaks-flagship-h100-chip-export-china-h800-2023-03-21/\nhttps://en.wikipedia.org/wiki/Hopper_(microarchitecture)\nhttps://www.cnbc.com/2023/10/17/us-bans-export-of-more-ai-chips-including-nvidia-h800-to-china.html\nhttps://thenextweb.com/news/nvidia-huang-deepseek-huawei-chips-horrible-outcome", "size": [5.24, 5.24], "position": { "x": 65.44, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-ch-h20", "name": "H20", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "GPU", "Date": "2023-11-24", "Milestone": false, "Impact score": 45 }, "description": "Full name: Nvidia H20. The H20 is the most powerful of three China-focused chips that Nvidia built to comply with the US export rules of 17 October 2023; the others were the L20 and the L2. Reuters reported on 24 November 2023 that Nvidia had told Chinese customers it was delaying the H20 to the first quarter of 2024 because server makers had trouble integrating it. Leaked specifications cited by TrendForce put its compute at about 15% of the H100's, with 96 GB of memory. The H20 became Nvidia's main China product; in April 2025 the US required licences for it, and Nvidia said it expected a charge of up to about $5.5 billion.\n\nhttps://www.businessday.co.za/bd/companies/telecoms-and-technology/2023-11-24-nvidia-to-delay-launch-of-ai-chip-in-china-say-sources/\nhttps://www.trendforce.com/news/2024/01/04/news-nvidias-china-exclusive-chip-h20-to-begin-production-in-q2-by-wistron/\nhttps://en.wikipedia.org/wiki/Hopper_(microarchitecture)", "size": [5.24, 5.24], "position": { "x": 118.757, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-blackwell", "name": "Blackwell", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "GPU", "Date": "2024-03-18", "Milestone": true, "Impact score": 70 }, "description": "Full name: Nvidia Blackwell. Blackwell is Nvidia's data-centre GPU architecture, announced at GTC on 18 March 2024 as the successor to Hopper. The B200 joins two dies of 104 billion transistors each, 208 billion in total, linked at 10 terabytes per second, with up to 192 GB of HBM3E memory; the GB200 pairs two GPUs with a Grace CPU and the NVL72 rack links 72 GPUs. Jensen Huang called it a processor for the generative-AI era, and Nvidia cited endorsements from the heads of Google, Meta, Microsoft, OpenAI and Oracle. In October 2024 a design flaw was reported and fixed with TSMC, and in November 2024 Morgan Stanley reported that 2025 Blackwell production was already sold out.\n\nhttps://nvidianews.nvidia.com/news/nvidia-blackwell-platform-arrives-to-power-a-new-era-of-computing\nhttps://www.reuters.com/technology/artificial-intelligence/nvidias-design-flaw-with-blackwell-ai-chips-now-fixed-ceo-says-2024-10-23/\nhttps://fortune.com/2024/11/12/jensen-huang-nvidia-ceo-leadership-mpp/\nhttps://en.wikipedia.org/wiki/Blackwell_(microarchitecture)", "size": [6.16, 6.16], "position": { "x": 138.573, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-ch-ascend-910c", "name": "Ascend 910C", "type": "Chip", "properties": { "Field": "Compute", "Chip kind": "AI accelerator", "Date": "2025-04-21", "Milestone": false, "Impact score": 38 }, "description": "The Ascend 910C is Huawei's AI accelerator, made by combining two Ascend 910B processors in one package, with its compute chiplets fabricated by SMIC on a 7 nm-class process. It became China's leading domestic answer to US export controls after the US required licences for Nvidia's H20 in April 2025: Reuters reported that Huawei planned mass shipments of the 910C as early as the following month, and Chinese cloud companies looked for substitutes. DeepSeek researchers put its inference performance at about 60% of the H100's, and analysts such as SemiAnalysis noted its high-bandwidth memory comes from Korea and its main wafers from TSMC. Wikipedia describes it as an evolution rather than a breakthrough.\n\nhttps://techblog.comsoc.org/2025/04/21/u-s-export-controls-on-nvidia-h20-ai-chips-enables-huaweis-910c-gpu-to-be-favored-by-ai-tech-giants-in-china/\nhttps://thenextweb.com/news/nvidia-huang-deepseek-huawei-chips-horrible-outcome\nhttps://en.wikipedia.org/wiki/HiSilicon", "size": [4.9, 4.9], "position": { "x": 206.162, "y": 0.23, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-dartmouth-workshop", "name": "Dartmouth workshop", "type": "Event", "properties": { "Field": "Symbolic AI", "Event kind": "Workshop or conference", "Date": "1956-06-18", "Milestone": true, "Impact score": 75 }, "description": "The Dartmouth Summer Research Project on Artificial Intelligence was a summer workshop at Dartmouth College that gave the field its name. John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon wrote the funding proposal (dated 31 August 1955 in the proposal's own citation) for the Rockefeller Foundation; McCarthy chose the term 'artificial intelligence', partly for its neutrality. About eleven people were planned, but attendance came and went: Ray Solomonoff's notes list twenty attendees and suggest the meeting ran from about 18 June to 17 August 1956, longer than the six weeks usually quoted, with only Solomonoff, Minsky and McCarthy staying throughout. It was not a directed project, yet it is credited with encouraging symbolic methods and limited-domain systems, and Allen Newell and Herbert Simon presented their Logic Theorist there to a lukewarm reception.\n\nhttps://en.wikipedia.org/wiki/Dartmouth_workshop\nhttps://spectrum.ieee.org/dartmouth-ai-workshop\nhttp://jmc.stanford.edu/articles/dartmouth/dartmouth.pdf\nPicture: document, A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, by John McCarthy et al, 1955.pdf, John McCarthy et al, Public domain, https://commons.wikimedia.org/wiki/File:A_Proposal_for_the_Dartmouth_Summer_Research_Project_on_Artificial_Intelligence,_by_John_McCarthy_et_al,_1955.pdf", "size": 5.25, "position": { "x": -270.06, "y": 31.92, "z": 0 }, "images": ["f4ad597f-7ad8-43c0-930f-2449cea14fef"] }, { "ID": "ai-hist-ev-lisp-machine-collapse", "name": "Lisp machine collapse", "type": "Event", "properties": { "Field": "Symbolic AI", "Event kind": "Corporate crisis", "Date": "1987-01-01", "Milestone": false, "Impact score": 38 }, "description": "In 1987 the market for Lisp machines, the specialised computers on which AI labs and expert-system companies ran their software, collapsed. Cheaper desktop machines such as Sun workstations, Apple and IBM computers could by then run Lisp programs as fast as the dedicated hardware; Wikipedia's AI winter article says an industry 'worth half a billion dollars' was replaced in a single year. Most Lisp machine makers, among them Symbolics and Lisp Machines Inc., had gone out of business by the early 1990s, and the collapse together with the cuts to DARPA's Strategic Computing programme is used to mark the start of the second AI winter.\n\nhttps://en.wikipedia.org/wiki/AI_winter\nhttps://en.wikipedia.org/wiki/Lisp_machine", "size": 3.69, "position": { "x": -201.693, "y": 31.92, "z": 0 } }, { "ID": "ai-hist-ev-deep-blue-match-1997", "name": "Deep Blue v Kasparov", "type": "Event", "properties": { "Field": "Games", "Event kind": "Match", "Date": "1997-05-03", "Milestone": false, "Impact score": 55 }, "description": "The 1997 rematch between Garry Kasparov and IBM's Deep Blue was played in six games from 3 to 11 May 1997 at the Equitable Center in New York City. Kasparov had won the first match 4 to 2 in Philadelphia in February 1996. In the rematch he lost Game 2 on 4 May, after which he accused IBM of cheating, and the match stood level at 2.5 each before Game 6, which Deep Blue won on 11 May in under twenty moves after a knight sacrifice, for a final score of 3.5 to 2.5. It was the first match victory by a computer over a reigning world champion under tournament regulations and became the defining public moment of AI before the deep-learning era.\n\nhttps://en.wikipedia.org/wiki/Deep_Blue_versus_Garry_Kasparov\nhttps://www.koreajoongangdaily.com/opinion/deep-blue-versus-kasparov/12607345", "size": 4.51, "position": { "x": -192.893, "y": 31.92, "z": 0 } }, { "ID": "ai-hist-ev-darpa-grand-challenge-2005", "name": "DARPA Grand Challenge 2005", "type": "Event", "properties": { "Field": "Robotics and driving", "Event kind": "Competition", "Date": "2005-10-08", "Milestone": true, "Impact score": 55 }, "description": "The second DARPA Grand Challenge was a race for driverless vehicles held on 8 October 2005 over a 212-kilometre (132-mile) desert course near the California and Nevada state line, for a prize of 2 million dollars. Of 23 vehicles, five completed the course; Stanford's Stanley won in about 6 hours 53 minutes (sources differ on the exact time and course length, 210 to 212 km), ahead of Carnegie Mellon's Sandstorm and H1ghlander. In the first challenge in 2004 the best vehicle had covered only about 11.8 kilometres. DARPA director Tony Tether said the agency had completed its mission and that a similar event was unlikely in 2006, and Sebastian Thrun predicted autonomous vehicles in everyday life within 50 years, perhaps 30.\n\nhttps://en.wikipedia.org/wiki/DARPA_Grand_Challenge_(2005)\nhttps://www.bioedonline.org/news/nature-news-archive/robot-car-scoops-us2-million-prize/\nhttps://www.aljazeera.com/news/2005/10/11/robocar-set-to-do-what-the-wheel-did", "size": 4.51, "position": { "x": -183.104, "y": 20.07, "z": 0 }, "images": ["688548e3-e46b-4062-adc4-a591b0a0a700"] }, { "ID": "ai-hist-ev-jeopardy-2011", "name": "Watson wins Jeopardy!", "type": "Event", "properties": { "Field": "Language", "Event kind": "Match", "Date": "2011-02-14", "Milestone": true, "Impact score": 65 }, "description": "IBM's Watson played the quiz show Jeopardy! against its two most successful champions, Ken Jennings and Brad Rutter, in a two-game match that aired from 14 to 16 February 2011 (it was taped in January at IBM's Watson Research Center in Yorktown Heights). Watson finished with 77,147 dollars against 24,000 for Jennings and 21,600 for Rutter, and won the 1 million dollar first prize, which IBM said it would donate to charity, split between World Vision and World Community Grid. Jennings wrote on his final screen 'I for one welcome our new computer overlords'. The demonstration showed that a machine could answer open-domain natural-language questions at human-champion level, and IBM promoted it as the start of commercial applications of Watson, notably in health care.\n\nhttps://en.wikipedia.org/wiki/Watson_(computer)\nhttps://www.discovermagazine.com/technology/what-is-champion-ibms-watson-seals-its-jeopardy-victory", "size": 4.9, "position": { "x": -156.913, "y": 31.92, "z": 0 } }, { "ID": "ai-hist-ev-ilsvrc-2012", "name": "ImageNet challenge 2012", "type": "Event", "properties": { "Field": "Vision", "Event kind": "Benchmark result", "Date": "2012-09-30", "Milestone": false, "Impact score": 65 }, "description": "The 2012 ImageNet Large Scale Visual Recognition Challenge, held in conjunction with the PASCAL Visual Object Classes Challenge 2012, asked teams to classify images into 1,000 categories. The University of Toronto team SuperVision, using a deep convolutional network, had the best top-5 error: 15.3 percent when it used extra ImageNet data from the Fall 2011 release and 16.4 percent using only the supplied training data. The best other entry, from ISI at the University of Tokyo, scored 26.2 percent with a weighted combination of conventional classifiers built on hand-designed features. The size of the gap, more than ten points, is what turned a conference result into a turning point and is commonly used to mark the beginning of the deep learning era.\n\nhttps://image-net.org/challenges/LSVRC/2012/results.html\nhttps://en.wikipedia.org/wiki/ImageNet", "size": 4.9, "position": { "x": -134.395, "y": 31.92, "z": 0 } }, { "ID": "ai-a-alphago-lee-sedol", "name": "AlphaGo v Lee Sedol", "type": "Event", "properties": { "Field": "Games", "Event kind": "Match", "Date": "2016-03-09", "Milestone": true, "Impact score": 75 }, "description": "The five-game Go match between Google DeepMind's AlphaGo and the 9-dan professional Lee Sedol was played from 9 to 15 March 2016 at the Four Seasons Hotel in Seoul, with a prize of 1 million dollars, which DeepMind said it would give to charities including UNICEF. AlphaGo won 4 to 1; all games ended by resignation. Michael Redmond called AlphaGo's move 37 in Game 2 'creative' and 'unique'; Lee won Game 4 after the move 78 that Gu Li called a 'divine move'; and Lee said after Game 3 that he had 'misjudged the capabilities of AlphaGo'. Lee received 170,000 dollars, the Korea Baduk Association gave AlphaGo an honorary 9 dan, and the match was widely compared with Deep Blue's win in 1997. Most observers had expected Lee to win.\n\nhttps://en.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol\nhttps://www.koreatimes.co.kr/www/news/tech/2016/03/133_200267.html\nPicture: diagram, Lee Sedol (B) vs AlphaGo (W) - Game 4.jpg, Wesalius, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Lee_Sedol_(B)_vs_AlphaGo_(W)_-_Game_4.jpg", "size": 5.25, "position": { "x": -96.348, "y": 20.07, "z": 0 }, "images": ["45b86ac6-8a3f-4834-b8cd-7ef082465a15"] }, { "ID": "ai-scale-ev-ke-jie", "name": "AlphaGo v Ke Jie", "type": "Event", "properties": { "Field": "Games", "Event kind": "Match", "Date": "2017-05-23", "Milestone": false, "Impact score": 50 }, "description": "At the Future of Go Summit in Wuzhen, China, on 23, 25 and 27 May 2017, DeepMind's AlphaGo Master, running on four TPUs on one machine, beat the world number one Ke Jie 3-0 in a three-game match. China's state broadcaster CCTV cancelled its plan to broadcast the games live, several Chinese websites cut off their streams and media were told to play down Google's role, according to Wikipedia; the only official live broadcast was on YouTube, which is blocked in China. DeepMind then announced that AlphaGo would retire and its Go team would turn to other AI problems. Kai-Fu Lee called the match China's Sputnik moment, and the State Council issued its national AI plan two months later.\n\nhttps://en.wikipedia.org/wiki/Future_of_Go_Summit\nhttps://digichina.stanford.edu/?p=826", "size": 4.29, "position": { "x": -71.247, "y": 31.92, "z": 0 } }, { "ID": "ai-scale-ev-openai-lp", "name": "OpenAI capped profit", "type": "Event", "properties": { "Field": "Agents and reasoning", "Event kind": "Corporate restructuring", "Date": "2019-03-11", "Milestone": false, "Impact score": 45 }, "description": "On 11 March 2019 OpenAI, founded in 2015 as a non-profit, created OpenAI LP, a 'capped' for-profit company in which investor returns were limited to 100 times the amount invested, as TechCrunch reported that day. The change let OpenAI raise the money that large-scale training demanded while the non-profit parent kept control of the mission. Within months Microsoft became its first large investor, and MIT Technology Review later argued that the move, together with the exclusive GPT-3 licence, showed a lab drifting from its founding promise to benefit humanity.\n\nhttps://en.wikipedia.org/wiki/OpenAI\nhttps://techcrunch.com/2019/03/11/openai-shifts-from-nonprofit-to-capped-profit-to-attract-capital/\nhttps://www.technologyreview.com/2020/09/23/1008729/openai-is-giving-microsoft-exclusive-access-to-its-gpt-3-language-model/", "size": 4.06, "position": { "x": -40.335, "y": 20.07, "z": 0 } }, { "ID": "ai-a-turing-award-2018", "name": "Turing Award 2018", "type": "Event", "properties": { "Field": "Neural networks", "Event kind": "Award or recognition", "Date": "2019-03-27", "Milestone": false, "Impact score": 45 }, "description": "The ACM A. M. Turing Award for 2018 went jointly to Yoshua Bengio, Geoffrey Hinton and Yann LeCun, with a citation for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing, according to Wikipedia's list of laureates. ACM announced it on 27 March 2019. The prize carries $1 million with financial support from Google. The three had kept neural-network research alive through the winters, and the award marked the field's acceptance by computer science's highest honour seven years after AlexNet.\n\nhttps://en.wikipedia.org/wiki/Turing_Award\nhttps://awards.acm.org/about/2018-turing", "size": 4.06, "position": { "x": -36.007, "y": -16.14, "z": 0 } }, { "ID": "ai-scale-ev-casp14", "name": "AlphaFold at CASP14", "type": "Event", "properties": { "Field": "Science", "Event kind": "Benchmark result", "Date": "2020-11-30", "Milestone": false, "Impact score": 45 }, "description": "CASP, the Critical Assessment of protein Structure Prediction, is a biennial blind contest in which research groups predict protein structures that experimenters have solved but not yet published. In the November 2020 round, CASP14, DeepMind's AlphaFold 2 reached a median score of 92.4 out of 100 on the GDT scale, which the organisers treat as roughly comparable to experimental accuracy. John Moult, who leads the team that runs CASP, told MIT Technology Review that it was the first use of AI to solve a serious problem, and David Baker of the University of Washington compared the result to AlphaGo's victory. The assessment is the moment structural biologists accepted that computation could rival experiment.\n\nhttps://www.technologyreview.com/2020/11/30/1012712/deepmind-protein-folding-ai-solved-biology-science-drugs-disease/\nhttps://en.wikipedia.org/wiki/AlphaFold", "size": 4.06, "position": { "x": 0.638, "y": 20.07, "z": 0 } }, { "ID": "ai-scale-ev-colorado-fair", "name": "AI art wins state fair", "type": "Event", "properties": { "Field": "Generative media", "Event kind": "Competition", "Date": "2022-08-29", "Milestone": false, "Impact score": 50 }, "description": "On 29 August 2022 Jason Allen's image Theatre D'opera Spatial won first place in the digital arts category for emerging artists at the Colorado State Fair, taking a $300 prize. Allen made it with Midjourney, using at least 624 text prompts and revisions, then edited it in Photoshop and upscaled it. The two judges said they had not known that Midjourney used AI but would have given him the top prize anyway. When Allen posted the result online, artists protested that the entry was not art, and the state's agriculture department compared it to a banana duct-taped to a wall; from 2023 the fair required entrants to say whether they used AI. In September 2023 the US Copyright Office review board found the work ineligible for copyright because the human input was de minimis.\n\nhttps://en.wikipedia.org/wiki/Th%C3%A9%C3%A2tre_D%27op%C3%A9ra_Spatial\nhttps://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html\nPicture: AI-generated image, Théâtre D’opéra Spatial.jpg, Jason M. Allen, Public domain, https://commons.wikimedia.org/wiki/File:Th%C3%A9%C3%A2tre_D%E2%80%99op%C3%A9ra_Spatial.jpg", "size": 4.29, "position": { "x": 33.813, "y": 20.07, "z": 0 }, "images": ["57ebc354-f447-448b-9ae1-9a58eab37037"] }, { "ID": "ai-a-bletchley", "name": "Bletchley AI summit", "type": "Event", "properties": { "Field": "Language", "Event kind": "Summit", "Date": "2023-11-01", "Milestone": false, "Impact score": 60 }, "description": "The AI Safety Summit was held on 1 and 2 November 2023 at Bletchley Park, the Second World War codebreaking site in Milton Keynes, hosted by UK Prime Minister Rishi Sunak. Attendees included US Vice President Kamala Harris, Commerce Secretary Gina Raimondo, European Commission President Ursula von der Leyen, Italian Prime Minister Giorgia Meloni, a Chinese vice-minister of science and technology, and technology leaders such as Elon Musk and Sam Altman. The summit produced the Bletchley Declaration on frontier AI risk, the commissioning of an International AI Safety Report, the official launch of the UK AI Safety Institute on 2 November, and a live conversation between Musk and Sunak on X. It started a series of summits in Seoul (2024), Paris (2025) and New Delhi (2026).\n\nhttps://en.wikipedia.org/wiki/AI_Safety_Summit\nhttps://www.bbc.com/news/uk-england-beds-bucks-herts-67273099\nPicture: UK Government hosts AI Summit at Bletchley Park (53302263013).jpg, UK Government, CC BY 2.0, https://commons.wikimedia.org/wiki/File:UK_Government_hosts_AI_Summit_at_Bletchley_Park_(53302263013).jpg", "size": 4.71, "position": { "x": 112.157, "y": -45.29, "z": 0 }, "images": ["75b004db-8cc1-4c11-8c17-e31b9bc3f181"] }, { "ID": "ai-a-altman-ouster", "name": "Altman ouster", "type": "Event", "properties": { "Field": "Language", "Event kind": "Corporate crisis", "Date": "2023-11-17", "Milestone": false, "Impact score": 55 }, "description": "On 17 November 2023 OpenAI's board removed Sam Altman as chief executive, saying he had not been consistently candid with it; chief technology officer Mira Murati became interim chief executive and the board named Emmett Shear on 19 November. On 20 November Microsoft announced it would hire Altman and colleagues for a new research team, and 745 of OpenAI's 770 employees signed a letter threatening to resign unless the board stepped down; co-founder and board member Ilya Sutskever signed it and publicly apologised. On 22 November Altman was reinstated with Microsoft chief executive Satya Nadella mediating, and a new interim board of Bret Taylor, Adam D'Angelo and Lawrence Summers was installed. An outside review published in March 2024 found the dismissal had not been required by safety or finance concerns. The episode showed how much of the AI boom rested on one company's governance and its dependence on Microsoft.\n\nhttps://en.wikipedia.org/wiki/Removal_of_Sam_Altman_from_OpenAI\nhttps://www.theverge.com/2023/11/22/23967223/sam-altman-returns-ceo-open-ai", "size": 4.51, "position": { "x": 116.989, "y": 20.07, "z": 0 } }, { "ID": "ai-a-nobel-physics-2024", "name": "Nobel Physics 2024", "type": "Event", "properties": { "Field": "Neural networks", "Event kind": "Award or recognition", "Date": "2024-10-08", "Milestone": true, "Impact score": 65 }, "description": "The Royal Swedish Academy of Sciences announced on 8 October 2024 that the Nobel Prize in Physics would go to John Hopfield and Geoffrey Hinton 'for foundational discoveries and inventions that enable machine learning with artificial neural networks'. Hopfield's 1982 network is a content-addressable memory of binary neurons; Hinton's Boltzmann machine work was cited. The award was the first Nobel for foundational AI research, and both laureates used the occasion to warn: Hopfield said he was deeply uneasy about the progress of AI and compared it to nuclear fission, and Hinton called for urgent safety research. Gizmodo headlined the coverage 'Nobel Prize Goes to Godfathers of AI Who Now Fear Their Work Is Growing Too Powerful'.\n\nhttps://en.wikipedia.org/wiki/John_Hopfield\nhttps://gizmodo.com/nobel-prize-goes-to-godfathers-of-ai-who-now-fear-their-work-is-growing-too-powerful-2000509098", "size": 4.9, "position": { "x": 164.494, "y": -16.14, "z": 0 } }, { "ID": "ai-a-nobel-chemistry-2024", "name": "Nobel Chemistry 2024", "type": "Event", "properties": { "Field": "Science", "Event kind": "Award or recognition", "Date": "2024-10-09", "Milestone": true, "Impact score": 60 }, "description": "The 2024 Nobel Prize in Chemistry, announced on 9 October 2024, was divided: one half to David Baker for computational protein design and the other half jointly to Demis Hassabis and John Jumper of Google DeepMind for protein structure prediction, according to Wikipedia's article on Jumper citing the Nobel Prize Outreach pages. The prediction work is AlphaFold 2, which won the CASP14 assessment in November 2020 and whose database has been used by researchers worldwide. The prize, a day after the physics prize for Hopfield and Hinton, made 2024 the year in which the Nobel committees recognised AI as a scientific tool.\n\nhttps://en.wikipedia.org/wiki/John_M._Jumper\nhttps://www.nobelprize.org/prizes/chemistry/2024/press-release/", "size": 4.71, "position": { "x": 174.704, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-ev-paris-summit", "name": "Paris AI summit", "type": "Event", "properties": { "Field": "Language", "Event kind": "Summit", "Date": "2025-02-10", "Milestone": false, "Impact score": 40 }, "description": "The AI Action Summit was held on 10 and 11 February 2025 at the Grand Palais in Paris, co-chaired by Emmanuel Macron and Narendra Modi, with over 1,000 participants from more than 100 countries according to Wikipedia. Fifty-eight countries, including France, China and India, signed its Statement on Inclusive and Sustainable Artificial Intelligence for People and the Planet; the United States and United Kingdom declined. US Vice President JD Vance warned against 'excessive regulation of the AI' and against cooperating with 'authoritarian regimes'. Investment pledges included the European Union's InvestAI plan of 200 billion euros and nearly 110 billion euros of private pledges in France (Reuters, 10 February 2025). The summit marked a shift from the safety framing of Bletchley and Seoul to an action and investment framing.\n\nhttps://en.wikipedia.org/wiki/AI_Action_Summit\nhttps://www.theguardian.com/technology/2025/feb/11/us-uk-paris-ai-summit-artificial-intelligence-declaration\nPicture: Ursula von der Leyen speaking during the Artificial Intelligence Action Summit at the Grand Palais, Paris (2025).jpg, Dati Bendo / European Union, 2025 / EC - Audiovisual Service, CC BY 4.0, https://commons.wikimedia.org/wiki/File:Ursula_von_der_Leyen_speaking_during_the_Artificial_Intelligence_Action_Summit_at_the_Grand_Palais,_Paris_(2025).jpg", "size": 3.8, "position": { "x": 195.12, "y": -36.18, "z": 0 }, "images": ["d80ee8cd-32aa-49d9-a217-d1ada62c47fb"] }, { "ID": "ai-boom-ev-imo-gold-2025", "name": "IMO gold for AI", "type": "Event", "properties": { "Field": "Agents and reasoning", "Event kind": "Benchmark result", "Date": "2025-07-19", "Milestone": true, "Impact score": 60 }, "description": "At the International Mathematical Olympiad of July 2025 both an OpenAI experimental model and a Google DeepMind Gemini Deep Think model reported solving five of six problems for 35 of 42 points, the gold-medal threshold, working in natural language within the 4.5-hour limit. Only 67 of 630 human contestants won gold. Google's result was graded and certified by IMO coordinators and announced on 21 July; OpenAI's model was not officially entered, announced its result on 19 July, and its score was disputed by a Google DeepMind researcher, who argued a point deduction would make it silver, a claim that is contested because the marking guideline is not public. The results were the first time general language models matched top human performance on Olympiad proofs and were widely cited as a sign of fast progress in reasoning models.\n\nhttps://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/\nhttps://www.axios.com/2025/07/21/openai-deepmind-math-olympiad-ai", "size": 4.71, "position": { "x": 222.32, "y": 20.07, "z": 0 } }, { "ID": "ai-p-alan-turing", "name": "Alan Turing", "type": "Person", "properties": { "Country": "United Kingdom", "Role": "Mathematician; author of Computing Machinery and Intelligence (1950)", "Impact score": 58 }, "description": "Alan Turing (1912 to 1954) was an English mathematician, logician and cryptanalyst, often called the father of theoretical computer science. His 1936 paper 'On Computable Numbers' introduced the Turing machine, and his 1950 paper 'Computing Machinery and Intelligence' proposed the Turing test as a way to judge whether a machine can be called intelligent. In 1952 he was convicted of gross indecency and lost his security clearance; the Queen granted him a pardon in 2013. The Turing Award of the Association for Computing Machinery, the highest honour in computing, is named after him, and his 1950 paper is where most histories of AI begin.\n\nhttps://en.wikipedia.org/wiki/Alan_Turing\nhttps://en.wikipedia.org/wiki/Computing_Machinery_and_Intelligence", "size": 3.54, "position": { "x": -286.765, "y": 58.46, "z": 0 }, "images": ["496774be-c1a1-46c3-b08b-233245001812"], "nodeImage": "496774be-c1a1-46c3-b08b-233245001812" }, { "ID": "ai-p-alec-radford", "name": "Alec Radford", "type": "Person", "properties": { "Country": "United States", "Role": "Researcher at OpenAI; lead author of GPT-1 and CLIP", "Impact score": 58 }, "description": "Alec Radford is the lead author of the first GPT paper (2018) and a central researcher behind OpenAI's language and vision models. He is first author of the CLIP paper 'Learning Transferable Visual Models From Natural Language Supervision' (26 February 2021), which trained on 400 million image-text pairs, and he is among the authors of the January 2020 scaling-laws paper. GPT-1, GPT-2, GPT-3 and CLIP each took the same idea, pretraining on very large unlabelled data from the internet, to a new scale, and his name is on the papers that defined that line.\n\nhttps://openai.com/research/language-unsupervised\nhttps://arxiv.org/abs/2103.00020\nhttps://arxiv.org/abs/2001.08361", "size": 3.54, "position": { "x": -49.27, "y": 58.46, "z": 0 } }, { "ID": "ai-p-alex-krizhevsky", "name": "Alex Krizhevsky", "type": "Person", "properties": { "Country": "Canada", "Role": "Lead author of AlexNet", "Impact score": 50 }, "description": "Alex Krizhevsky is a Canadian computer scientist who did his PhD at the University of Toronto under Geoffrey Hinton. He co-developed AlexNet with Ilya Sutskever and Hinton, using two GeForce GPUs, and the network won the ImageNet challenge and was presented at NeurIPS in 2012. Shortly afterwards he and Sutskever sold their start-up DNNresearch to Google; he left Google in September 2017 and joined Dessa. He is the main author of the CIFAR-10 and CIFAR-100 datasets.\n\nhttps://en.wikipedia.org/wiki/Alex_Krizhevsky", "size": 3.3, "position": { "x": -148.754, "y": 58.46, "z": 0 } }, { "ID": "ai-hist-p-allen-newell", "name": "Allen Newell", "type": "Person", "properties": { "Country": "United States", "Role": "Co-creator of the Logic Theorist; professor at Carnegie Mellon", "Impact score": 50 }, "description": "Allen Newell (1927 to 1992) was an American researcher who joined the RAND Corporation in 1950 and earned his doctorate at Carnegie Mellon under Herbert Simon. With Simon and Cliff Shaw he created the Logic Theorist in 1956, often considered the first true AI program, and in 1957 the General Problem Solver, built on means-ends analysis. They presented the Logic Theorist at the Dartmouth workshop. Newell and Simon shared the 1975 Turing Award, and Newell later developed the Soar cognitive architecture, tied to his 1990 unified theory of cognition.\n\nhttps://en.wikipedia.org/wiki/Allen_Newell\nhttps://en.wikipedia.org/wiki/Logic_Theorist", "size": 3.3, "position": { "x": -279.325, "y": 58.46, "z": 0 } }, { "ID": "ai-p-andrew-ng", "name": "Andrew Ng", "type": "Person", "properties": { "Country": "United States", "Role": "Co-founder of Google Brain and Coursera; Stanford professor", "Impact score": 50 }, "description": "Andrew Ng (born 1976 in London) is a British-American computer scientist and entrepreneur. He joined Stanford as an assistant professor in 2002 and directed its AI Lab. In 2011 he founded the Google Brain project while at Google, in 2012 co-founded the online-learning company Coursera with Daphne Koller, and from 2014 to his resignation in March 2017 was chief scientist at Baidu. In 2009 he co-authored with Rajat Raina and Anand Madhavan the paper on training deep belief networks on GPUs, and his Stanford group promoted GPUs for deep learning around 2008. After leaving Baidu he launched DeepLearning.AI.\n\nhttps://en.wikipedia.org/wiki/Andrew_Ng\nhttps://mlanthology.org/icml/2009/raina2009icml-large\nPicture: Andrew Ng at TechCrunch Disrupt SF 2017.jpg, TechCrunch, CC BY 2.0, https://commons.wikimedia.org/wiki/File:Andrew_Ng_at_TechCrunch_Disrupt_SF_2017.jpg", "size": 3.3, "position": { "x": -170.07, "y": 58.46, "z": 0 }, "images": ["d8c02bd9-ebcb-4c63-b6e5-36dcb15b32dc"], "nodeImage": "d8c02bd9-ebcb-4c63-b6e5-36dcb15b32dc" }, { "ID": "ai-p-ashish-vaswani", "name": "Ashish Vaswani", "type": "Person", "properties": { "Country": "United States", "Role": "First-listed author of the Transformer paper, Google, 2017", "Impact score": 40 }, "description": "Ashish Vaswani is the first-listed author of 'Attention Is All You Need', the 2017 paper from Google that introduced the transformer architecture. The paper credits all eight authors as equal contributors and states that the listing order was randomised, so his position says nothing about rank. He and his co-authors worked at Google when the paper appeared; the Wikipedia article on the paper notes that each of them later left to join other companies or to found startups. Because the transformer became the basis of nearly every large language and image model after 2017, the author list is one of the most cited groups in modern machine learning.\n\nhttps://en.wikipedia.org/wiki/Attention_Is_All_You_Need\nhttps://arxiv.org/abs/1706.03762", "size": 2.95, "position": { "x": -69.64, "y": 58.46, "z": 0 } }, { "ID": "ai-p-claude-shannon", "name": "Claude Shannon", "type": "Person", "properties": { "Country": "United States", "Role": "Information theorist; co-author of the Dartmouth proposal", "Impact score": 38 }, "description": "Claude Shannon (1916 to 2001) was an American mathematician and engineer, known as the father of information theory after his 1948 paper on communication. He worked at Bell Labs during the Second World War, taught at MIT from 1956 to 1978, and co-authored the 1955 proposal for the Dartmouth workshop with John McCarthy, Marvin Minsky and Nathaniel Rochester. His 1950 paper 'Programming a Computer for Playing Chess' helped launch computer chess, and the same year he built Theseus, an electromechanical mouse that learned to solve a maze by trial and error.\n\nhttps://en.wikipedia.org/wiki/Claude_Shannon\nhttps://en.wikipedia.org/wiki/Dartmouth_workshop\nPicture: C.E. Shannon. Tekniska museet 43069 (2x3 crop).jpg, unknown author, CC BY 2.0, https://commons.wikimedia.org/wiki/File:C.E._Shannon._Tekniska_museet_43069_(2x3_crop).jpg", "size": 2.87, "position": { "x": -265.355, "y": 58.46, "z": 0 }, "images": ["04c77487-4f6c-407e-97f1-5fa081784daf"], "nodeImage": "04c77487-4f6c-407e-97f1-5fa081784daf" }, { "ID": "ai-p-dario-amodei", "name": "Dario Amodei", "type": "Person", "properties": { "Country": "United States", "Role": "Co-founder and CEO of Anthropic; former VP of Research at OpenAI", "Impact score": 58 }, "description": "Dario Amodei was vice president of research at OpenAI and a co-author of the January 2020 scaling-laws paper before he and his sister Daniela left to found Anthropic in January 2021; the Wikipedia account says they left over directional differences. TechCrunch described him as Anthropic's chief executive when the company announced $124 million of funding on 28 May 2021. His move from OpenAI to Anthropic is the founding story of the best-known rival of OpenAI.\n\nhttps://en.wikipedia.org/wiki/Anthropic\nhttps://techcrunch.com/2021/05/28/anthropic-is-the-new-ai-research-outfit-from-openais-dario-amodei-and-it-has-124m-to-burn/\nhttps://arxiv.org/abs/2001.08361\nPicture: Dario Amodei at TechCrunch Disrupt 2023 01.jpg, TechCrunch, CC BY 2.0, https://commons.wikimedia.org/wiki/File:Dario_Amodei_at_TechCrunch_Disrupt_2023_01.jpg", "size": 3.54, "position": { "x": -62.55, "y": 58.46, "z": 0 }, "images": ["fe220818-bf9d-40b0-9f20-95beb751e4d9"], "nodeImage": "fe220818-bf9d-40b0-9f20-95beb751e4d9" }, { "ID": "ai-scale-p-david-holz", "name": "David Holz", "type": "Person", "properties": { "Country": "United States", "Role": "Founder of Midjourney; co-founder of Leap Motion", "Impact score": 38 }, "description": "David Holz is the founder of Midjourney, Inc. and its AI image generator, which opened its public beta on 12 July 2022 and which users reach through Discord. Before Midjourney he co-founded Leap Motion, a maker of hand-tracking devices. Midjourney, run by a small independent team, became a leading rival to OpenAI's DALL-E and to Stable Diffusion, and a Midjourney picture won a state art contest in August 2022.\n\nhttps://en.wikipedia.org/wiki/Midjourney\nhttps://www.theregister.com/2022/08/01/david_holz_midjourney/", "size": 2.87, "position": { "x": 29.324, "y": 58.46, "z": 0 } }, { "ID": "ai-hist-p-david-rumelhart", "name": "David Rumelhart", "type": "Person", "properties": { "Country": "United States", "Role": "Psychologist; co-author of the backpropagation paper", "Impact score": 40 }, "description": "David Rumelhart (1942 to 2011) was an American psychologist working on cognition and neural networks. He taught at the University of California, San Diego from 1967 to 1987 and then at Stanford until 1998. With James McClelland he edited the 1986 Parallel Distributed Processing volumes, a core text of cognitive science, and with Geoffrey Hinton and Ronald Williams he wrote the 1986 Nature paper that popularised backpropagation. The David E. Rumelhart Prize was created in his honour in 2000; he died in 2011 after Pick's disease disabled him.\n\nhttps://en.wikipedia.org/wiki/David_Rumelhart\nPicture: DavidRumelhart-IJCNNseattle1991-07-08.jpg, Rolf Kickuth, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:DavidRumelhart-IJCNNseattle1991-07-08.jpg", "size": 2.95, "position": { "x": -217.23, "y": 58.46, "z": 0 }, "images": ["0a0606d3-c742-4ee3-ab0b-1afd2f04c845"], "nodeImage": "0a0606d3-c742-4ee3-ab0b-1afd2f04c845" }, { "ID": "ai-hist-p-david-silver", "name": "David Silver", "type": "Person", "properties": { "Country": "United Kingdom", "Role": "Lead of the AlphaGo project at DeepMind", "Impact score": 50 }, "description": "David Silver is a researcher at DeepMind and a professor at University College London who leads reinforcement learning research. He led the AlphaGo project, the first program to defeat a top professional at full-size Go, developed AlphaZero, and co-authored the Nature paper on deep Q-networks, for which he was elected a Fellow of the Royal Society. He received the 2019 ACM Prize in Computing and co-wrote with Demis Hassabis the January 2016 announcement of AlphaGo's Nature paper.\n\nhttps://en.wikipedia.org/wiki/David_Silver_(computer_scientist)\nhttps://research.google/blog/alphago-mastering-the-ancient-game-of-go-with-machine-learning/", "size": 3.3, "position": { "x": -126.514, "y": 58.46, "z": 0 } }, { "ID": "ai-p-demis-hassabis", "name": "Demis Hassabis", "type": "Person", "properties": { "Country": "United Kingdom", "Role": "Co-founder and chief executive of DeepMind; Nobel Prize in Chemistry 2024", "Impact score": 82 }, "description": "Demis Hassabis (born 1976 in London) is a British AI researcher and entrepreneur, a chess prodigy who reached master standard at 13. He co-designed Theme Park at Bullfrog, was lead AI programmer on Black & White at Lionhead and founded the games studio Elixir (1998 to 2005), studied computer science at Cambridge and took a PhD in cognitive neuroscience at UCL in 2009. In 2010 he co-founded DeepMind with Shane Legg and Mustafa Suleyman and led it through the Deep Q-network and AlphaGo; Google bought it in 2014. He co-founded Isomorphic Labs, shared the 2024 Nobel Prize in Chemistry with John Jumper for AlphaFold, and was knighted in 2024.\n\nhttps://en.wikipedia.org/wiki/Demis_Hassabis\nPicture: Demis Hassabis, 2024 Nobel Prize Laureate in Chemistry 7 (cropped).jpg, John Sears, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Demis_Hassabis,_2024_Nobel_Prize_Laureate_in_Chemistry_7_(cropped).jpg", "size": 4.13, "position": { "x": -98.636, "y": 67.69, "z": 0 }, "images": ["48c20f55-5a3e-47a6-a46e-b535744057b9"], "nodeImage": "48c20f55-5a3e-47a6-a46e-b535744057b9" }, { "ID": "ai-p-donald-trump", "name": "Donald Trump", "type": "Person", "properties": { "Country": "United States", "Role": "President of the United States 2025-", "Impact score": 65 }, "description": "Donald Trump has been President of the United States since January 2025 (and also from 2017 to 2021). On his first day in office he revoked Biden's Executive Order 14110, on 23 January 2025 signed Executive Order 14179, on 21 January announced the Stargate infrastructure venture with Sam Altman, Larry Ellison and Masayoshi Son, and on 23 July 2025 published America's AI Action Plan with three executive orders. In February 2026 he ordered federal agencies to stop using Anthropic's technology after a dispute with the Pentagon, and in June 2026 his administration restricted and then restored access to Anthropic's Mythos-class models. He called DeepSeek's rise a 'wake up call' in January 2025, according to news reports.\n\nhttps://en.wikipedia.org/wiki/Executive_Order_14110\nhttps://fortune.com/2026/03/06/pentagon-officially-defines-anthropic-as-supply-chain-risk/", "size": 3.73, "position": { "x": 182.935, "y": 58.46, "z": 0 }, "images": ["c5b49088-a3f7-4cb1-b73c-16bcc8ad3e8f"], "nodeImage": "c5b49088-a3f7-4cb1-b73c-16bcc8ad3e8f" }, { "ID": "ai-p-elon-musk", "name": "Elon Musk", "type": "Person", "properties": { "Country": "United States", "Role": "Co-founder of OpenAI and founder of xAI", "Impact score": 65 }, "description": "Elon Musk is a business executive, chief executive and largest shareholder of Tesla and SpaceX. He co-founded OpenAI in December 2015 as a non-profit AI research company; Wikipedia states that he pledged 1 billion dollars but donated only 50 million, and that he left OpenAI's board in 2018. He launched xAI in July 2023 to build a generative AI program competing with ChatGPT. He was among the signatories of the January 2015 open letter on robust and beneficial AI.\n\nhttps://en.wikipedia.org/wiki/Elon_Musk\nhttps://en.wikipedia.org/wiki/Open_Letter_on_Artificial_Intelligence\nPicture: Elon Musk - 54820081119 (cropped).jpg, Gage Skidmore, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Elon_Musk_-_54820081119_(cropped).jpg", "size": 3.73, "position": { "x": -103.264, "y": 58.46, "z": 0 }, "images": ["d8a78def-9458-46ed-8315-26e57d8bc207"], "nodeImage": "d8a78def-9458-46ed-8315-26e57d8bc207" }, { "ID": "ai-p-emad-mostaque", "name": "Emad Mostaque", "type": "Person", "properties": { "Country": "United Kingdom", "Role": "Co-founder and CEO of Stability AI until March 2024", "Impact score": 45 }, "description": "Emad Mostaque is a British-Bangladeshi businessman and former hedge fund manager who co-founded Stability AI and was its chief executive when it released Stable Diffusion in August 2022. In defending the open release he argued that it was people's responsibility whether they operate the technology ethically, morally and legally. After a Forbes report in June 2023 alleged that he had exaggerated parts of his record, which he disputed, and months of investor pressure, he resigned as CEO and from the board on 23 March 2024 to pursue what he called decentralised AI.\n\nhttps://en.wikipedia.org/wiki/Emad_Mostaque\nhttps://en.wikipedia.org/wiki/Stable_Diffusion\nPicture: Mostaque.jpg, UK Government, CC BY 2.0, https://commons.wikimedia.org/wiki/File:Mostaque.jpg", "size": 3.13, "position": { "x": 70.203, "y": 58.46, "z": 0 }, "images": ["d767d39d-46e4-45b0-8897-cccdff0d66d6"], "nodeImage": "d767d39d-46e4-45b0-8897-cccdff0d66d6" }, { "ID": "ai-p-fei-fei-li", "name": "Fei-Fei Li", "type": "Person", "properties": { "Country": "United States", "Role": "Stanford professor; creator of ImageNet", "Impact score": 50 }, "description": "Fei-Fei Li (born 1976 in Beijing) is a professor of computer science at Stanford University, which she joined in 2009 and where she became a full professor in 2018. She began developing ImageNet in 2007 while at Princeton, and the dataset and its annual challenge became the benchmark that made deep learning's progress visible. From January 2017 to fall 2018 she was chief scientist of AI and machine learning at Google Cloud during a sabbatical, she is a founding co-director of Stanford's Human-Centered AI Institute, and in 2024 she co-founded World Labs to build 'spatial intelligence' AI.\n\nhttps://en.wikipedia.org/wiki/Fei-Fei_Li\nPicture: Fei-Fei Li at AI for Good 2017.jpg, ITU Pictures, CC BY 2.0, https://commons.wikimedia.org/wiki/File:Fei-Fei_Li_at_AI_for_Good_2017.jpg", "size": 3.3, "position": { "x": -162.87, "y": 58.46, "z": 0 }, "images": ["3b820221-7b66-44ec-a97d-f90b01c31fcd"], "nodeImage": "3b820221-7b66-44ec-a97d-f90b01c31fcd" }, { "ID": "ai-p-frank-rosenblatt", "name": "Frank Rosenblatt", "type": "Person", "properties": { "Country": "United States", "Role": "Inventor of the perceptron at Cornell", "Impact score": 48 }, "description": "Frank Rosenblatt (1928 to 1971) was an American psychologist and engineer who led the cognitive systems section at the Cornell Aeronautical Laboratory in Buffalo, where he developed the perceptron. He returned to Cornell in 1959 to direct its Cognitive Systems Research Program, completed the Mark I Perceptron in 1960 (now held by the Smithsonian) and summarised his work in the book Principles of Neurodynamics (1962). He died in a boating accident on Chesapeake Bay on his 43rd birthday. The IEEE awards an annual Frank Rosenblatt Award in his name; Cornell's own 2019 account argues that the principles behind his perceptron underlie modern deep learning.\n\nhttps://en.wikipedia.org/wiki/Frank_Rosenblatt\nhttps://news.cornell.edu/node/313946\nPicture: Frank Rosenblatt.jpg, unknown author, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Frank_Rosenblatt.jpg", "size": 3.23, "position": { "x": -269.826, "y": 67.69, "z": 0 }, "images": ["6483e7ec-6961-4e93-8957-393fd3d33f0a"], "nodeImage": "6483e7ec-6961-4e93-8957-393fd3d33f0a" }, { "ID": "ai-p-garry-kasparov", "name": "Garry Kasparov", "type": "Person", "properties": { "Country": "International", "Role": "World chess champion 1985 to 2000; opponent of Deep Blue", "Impact score": 60 }, "description": "Garry Kasparov (born 1963 in Baku, then in the Soviet Union, as Garik Weinstein) is a Russian chess grandmaster who also holds Croatian citizenship since 2014. He was World Chess Champion from 1985 to 2000, undisputed until 1993 and afterwards holder of the classical title until he lost it to Vladimir Kramnik. He beat IBM's Deep Blue 4 to 2 in 1996 and lost the 1997 rematch 3.5 to 2.5, the first loss of a reigning champion to a computer in a match, and accused IBM of cheating. He retired from regular competition in 2005 and chaired the Human Rights Foundation from 2011 to 2024.\n\nhttps://en.wikipedia.org/wiki/Garry_Kasparov\nPicture: Garri Kasparow (18776605665) (cropped) 2.jpg, Fryta 73 from Strzegom, CC BY-SA 2.0, https://commons.wikimedia.org/wiki/File:Garri_Kasparow_(18776605665)_(cropped)_2.jpg", "size": 3.59, "position": { "x": -192.09, "y": 58.46, "z": 0 }, "images": ["d397639c-4fec-4c9c-ab7e-f18d9fb5506c"], "nodeImage": "d397639c-4fec-4c9c-ab7e-f18d9fb5506c" }, { "ID": "ai-p-geoffrey-hinton", "name": "Geoffrey Hinton", "type": "Person", "properties": { "Country": "Canada", "Role": "Professor at Toronto; Turing Award 2018; Nobel Prize in Physics 2024", "Impact score": 100 }, "description": "Geoffrey Hinton (born 1947 in London) is a British-Canadian computer scientist, educated at Cambridge and with a PhD in AI from Edinburgh. He co-authored the 1986 Nature paper that popularised backpropagation with David Rumelhart and Ronald Williams, co-invented Boltzmann machines in 1985, and has been a professor at the University of Toronto since 1987. With his students Alex Krizhevsky and Ilya Sutskever he built AlexNet and in 2012 founded DNNresearch, which Google bought in 2013; he worked at Google until leaving in May 2023 to speak freely about the risks of AI. He shared the 2018 Turing Award with Yoshua Bengio and Yann LeCun and the 2024 Nobel Prize in Physics with John Hopfield.\n\nhttps://en.wikipedia.org/wiki/Geoffrey_Hinton\nPicture: Geoffrey Hinton in 2026.jpg, Cmichel67, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Geoffrey_Hinton_in_2026.jpg", "size": 4.5, "position": { "x": -209.18, "y": 58.46, "z": 0 }, "images": ["b8f91bca-5e78-4e35-8c39-d85b064fdaff"], "nodeImage": "b8f91bca-5e78-4e35-8c39-d85b064fdaff" }, { "ID": "ai-hist-p-greg-brockman", "name": "Greg Brockman", "type": "Person", "properties": { "Country": "United States", "Role": "Co-founder and president of OpenAI", "Impact score": 45 }, "description": "Greg Brockman (born 1987 in Thompson, North Dakota) joined the payments company Stripe in 2010 after leaving MIT, became its first chief technology officer in 2013 and left in May 2015. In December 2015 he co-founded OpenAI with Sam Altman, Ilya Sutskever and others and served as its chief technology officer; he is now its president. In November 2023 he was removed from the board when Sam Altman was fired and announced that he had quit, and returned to OpenAI after Altman was reinstated.\n\nhttps://en.wikipedia.org/wiki/Greg_Brockman\nhttps://techcrunch.com/2015/12/11/non-profit-openai-launches-with-backing-from-elon-musk-and-sam-altman\nPicture: Disrupt SF TechCrunch Disrupt San Francisco 2019 - Day 2 (48838200316) (cropped).jpg, TechCrunch, CC BY 2.0, https://commons.wikimedia.org/wiki/File:Disrupt_SF_TechCrunch_Disrupt_San_Francisco_2019_-_Day_2_(48838200316)_(cropped).jpg", "size": 3.13, "position": { "x": -88.744, "y": 58.46, "z": 0 }, "images": ["715d12af-ec8a-425f-ab78-8d9c26eb746c"], "nodeImage": "715d12af-ec8a-425f-ab78-8d9c26eb746c" }, { "ID": "ai-hist-p-herbert-simon", "name": "Herbert Simon", "type": "Person", "properties": { "Country": "United States", "Role": "Co-creator of the Logic Theorist; economist and psychologist", "Impact score": 50 }, "description": "Herbert A. Simon (1916 to 2001) was an American scholar who spent most of his career at Carnegie Mellon University, from 1949 to 2001. He created the Logic Theorist with Allen Newell in 1956, shared the 1975 Turing Award with Newell, and won the 1978 Nobel Memorial Prize in Economic Sciences for work on organisational decision-making; he is known for the concepts of bounded rationality and satisficing. In 1957 he predicted that a computer would match human chess players within ten years, a forecast that took about forty years to come true with Deep Blue.\n\nhttps://en.wikipedia.org/wiki/Herbert_A._Simon\nhttps://en.wikipedia.org/wiki/Logic_Theorist", "size": 3.3, "position": { "x": -272.125, "y": 58.46, "z": 0 }, "images": ["2b8cc1d4-3a17-46f0-a76c-97ef2bb6ce64"], "nodeImage": "2b8cc1d4-3a17-46f0-a76c-97ef2bb6ce64" }, { "ID": "ai-p-ian-goodfellow", "name": "Ian Goodfellow", "type": "Person", "properties": { "Country": "United States", "Role": "Inventor of generative adversarial networks", "Impact score": 48 }, "description": "Ian Goodfellow (born 1987) is an American computer scientist who earned bachelor's and master's degrees at Stanford and a PhD at the Université de Montréal, where Yoshua Bengio advised him. He is best known for inventing generative adversarial networks in 2014. He worked at Google Brain, was an early employee at OpenAI in 2016 and returned to Google in 2017; he became Apple's director of machine learning in 2019, left in 2022 over a return-to-office policy and then joined Google DeepMind. He co-wrote the textbook Deep Learning (2016).\n\nhttps://en.wikipedia.org/wiki/Ian_Goodfellow\nhttps://arxiv.org/abs/1406.2661\nPicture: Ian Goodfellow.jpg, Ian Goodfellow, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Ian_Goodfellow.jpg", "size": 3.23, "position": { "x": -122.523, "y": 67.69, "z": 0 }, "images": ["a59b7e5e-d306-4545-8e10-a336da1ccfab"], "nodeImage": "a59b7e5e-d306-4545-8e10-a336da1ccfab" }, { "ID": "ai-p-ilya-sutskever", "name": "Ilya Sutskever", "type": "Person", "properties": { "Country": "United States", "Role": "Co-creator of AlexNet; co-founder and former chief scientist of OpenAI", "Impact score": 75 }, "description": "Ilya Sutskever (born 1986 in Gorky, now Nizhny Novgorod, in the Soviet Union) is an Israeli and Canadian computer scientist. He earned a PhD at the University of Toronto in 2013 under Geoffrey Hinton, co-created AlexNet with Alex Krizhevsky and Hinton in 2012, and joined Google Brain in 2013 after Google bought their company DNNresearch, where he helped develop sequence-to-sequence learning. He co-founded OpenAI and served as its chief scientist from late 2015 until his departure in May 2024, and in June 2024 co-founded Safe Superintelligence Inc. with Daniel Gross and Daniel Levy, later becoming its CEO.\n\nhttps://en.wikipedia.org/wiki/Ilya_Sutskever\nPicture: Ilya Sutskever and Sam Altman in TAU (cropped).jpg, Eladkarmel, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Ilya_Sutskever_and_Sam_Altman_in_TAU_(cropped).jpg", "size": 3.97, "position": { "x": -140.884, "y": 58.46, "z": 0 }, "images": ["a6d5257b-1c8d-4a80-9cc8-81b49cba3d44"], "nodeImage": "a6d5257b-1c8d-4a80-9cc8-81b49cba3d44" }, { "ID": "ai-scale-p-jacob-devlin", "name": "Jacob Devlin", "type": "Person", "properties": { "Country": "United States", "Role": "First author of BERT, Google AI, 2018", "Impact score": 40 }, "description": "Jacob Devlin is the first author of the BERT paper, written with Ming-Wei Chang, Kenton Lee and Kristina Toutanova at Google AI and posted on 11 October 2018. The paper reported state-of-the-art results on eleven language-understanding tasks with a single pretrained transformer that could be fine-tuned cheaply; Google open-sourced the code and models in November 2018. He is also listed among the authors of Google's 2022 PaLM paper. BERT was later built into Google Search, where from October 2019 it helped interpret English queries.\n\nhttps://arxiv.org/abs/1810.04805\nhttps://arxiv.org/abs/2204.02311\nhttps://en.wikipedia.org/wiki/BERT_(language_model)", "size": 2.95, "position": { "x": -42.18, "y": 58.46, "z": 0 } }, { "ID": "ai-scale-p-jakob-uszkoreit", "name": "Jakob Uszkoreit", "type": "Person", "properties": { "Country": "United States", "Role": "Google researcher; co-author of the Transformer and Vision Transformer papers", "Impact score": 30 }, "description": "Jakob Uszkoreit was a Google researcher and co-author of 'Attention Is All You Need'. According to the account in the Wikipedia article on the paper, he liked the sound of the word 'Transformer' and so the architecture got its name; an early design document was titled 'Transformers: Iterative Self-Attention and Processing for Various Tasks' and the team called itself Team Transformer. On 31 August 2017 he wrote the Google Research blog post that explained the new architecture to a general audience, and in October 2020 he was a co-author of the Vision Transformer paper that applied the same design to images.\n\nhttps://en.wikipedia.org/wiki/Attention_Is_All_You_Need\nhttps://research.google/blog/transformer-a-novel-neural-network-architecture-for-language-understanding/\nhttps://arxiv.org/abs/2010.11929", "size": 2.5, "position": { "x": -55.91, "y": 58.46, "z": 0 } }, { "ID": "ai-p-james-lighthill", "name": "James Lighthill", "type": "Person", "properties": { "Country": "United Kingdom", "Role": "Cambridge mathematician; author of the 1973 report on AI", "Impact score": 40 }, "description": "Sir Michael James Lighthill (1924 to 1998) was a British applied mathematician known for fluid dynamics and aeroacoustics, which he founded as a field; he formulated Lighthill's eighth power law on jet-engine noise. He was Lucasian Professor of Mathematics at Cambridge from 1969 to 1979 and was knighted in 1971. In 1972 the Science Research Council asked him to review academic AI research; his critical 1973 report is associated with the onset of the first AI winter in Britain.\n\nhttps://en.wikipedia.org/wiki/James_Lighthill\nhttps://en.wikipedia.org/wiki/Lighthill_report", "size": 2.95, "position": { "x": -245.255, "y": 58.46, "z": 0 } }, { "ID": "ai-p-jensen-huang", "name": "Jensen Huang", "type": "Person", "properties": { "Country": "United States", "Role": "Co-founder and CEO of Nvidia", "Impact score": 72 }, "description": "Jensen Huang is the founder and chief executive of Nvidia and the public face of the company's turn from graphics to AI computing. He announced the V100 in 2017, the A100 in 2020, the H100 in 2022 and Blackwell at GTC on 18 March 2024, calling it a processor for the generative AI era. In May 2023 he said that the computer industry was going through two simultaneous transitions, accelerated computing and generative AI, as Nvidia forecast revenue far above expectations. In April 2026 he told the Dwarkesh podcast that it would be a horrible outcome for the US if DeepSeek optimised its models for Huawei's chips. Entrepreneur estimated his net worth at about $180 billion in October 2025, mostly from his roughly 3.5% stake in Nvidia.\n\nhttps://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2024\nhttps://en.wikipedia.org/wiki/Blackwell_(microarchitecture)\nhttps://thenextweb.com/news/nvidia-huang-deepseek-huawei-chips-horrible-outcome\nhttps://www.entrepreneur.com/business-news/nvidia-becomes-the-first-to-pass-5-trillion-in-market-value\nPicture: Jensen Huang (cropped).jpg, Prime Minister's Office, GODL-India, https://commons.wikimedia.org/wiki/File:Jensen_Huang_(cropped).jpg", "size": 3.9, "position": { "x": 138.573, "y": 58.46, "z": 0 }, "images": ["a9cabbea-1407-44f3-85a0-4166588bfbc7"], "nodeImage": "a9cabbea-1407-44f3-85a0-4166588bfbc7" }, { "ID": "ai-p-joe-biden", "name": "Joe Biden", "type": "Person", "properties": { "Country": "United States", "Role": "President of the United States 2021-2025", "Impact score": 50 }, "description": "Joe Biden was the 46th President of the United States, from January 2021 to January 2025. He signed Executive Order 14110 on safe, secure and trustworthy AI on 30 October 2023, the most comprehensive US governance action on AI to that date according to Wikipedia, and the order was timed two days before the Bletchley summit, where Vice President Kamala Harris represented the United States. The order was revoked by his successor Donald Trump on 20 January 2025, the first day of his term.\n\nhttps://en.wikipedia.org/wiki/Executive_Order_14110\nhttps://en.wikipedia.org/wiki/AI_Safety_Summit", "size": 3.3, "position": { "x": 36.094, "y": 58.46, "z": 0 }, "images": ["c9dc8049-95e4-44cd-89c0-50fa660915ca"], "nodeImage": "c9dc8049-95e4-44cd-89c0-50fa660915ca" }, { "ID": "ai-p-john-hopfield", "name": "John Hopfield", "type": "Person", "properties": { "Country": "United States", "Role": "Physicist; Hopfield network (1982); Nobel Prize in Physics 2024", "Impact score": 60 }, "description": "John Hopfield (born 1933 in Chicago) is an American physicist who taught at Princeton (1964 to 1980, and again from 1997) and Caltech (1980 to 1997) and worked at Bell Labs and Berkeley before. His 1982 paper in the Proceedings of the National Academy of Sciences introduced the network now named after him, an associative memory with an energy function. In 2024 he shared the Nobel Prize in Physics with Geoffrey Hinton for foundational discoveries and inventions that enable machine learning with artificial neural networks.\n\nhttps://en.wikipedia.org/wiki/John_Hopfield\nhttps://en.wikipedia.org/wiki/Hopfield_network\nPicture: John J. Hopfield delivering his lecture at the 2024 Nobel Prize Lectures in Stockholm.jpg, Jay Dixit, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:John_J._Hopfield_delivering_his_lecture_at_the_2024_Nobel_Prize_Lectures_in_Stockholm.jpg", "size": 3.59, "position": { "x": -226.654, "y": 58.46, "z": 0 }, "images": ["56667b1a-20f3-488f-af2c-6cb956522880"], "nodeImage": "56667b1a-20f3-488f-af2c-6cb956522880" }, { "ID": "ai-p-john-jumper", "name": "John Jumper", "type": "Person", "properties": { "Country": "United Kingdom", "Role": "DeepMind scientist; AlphaFold 2 lead; Nobel laureate in Chemistry 2024", "Impact score": 60 }, "description": "John Jumper led the DeepMind team that built AlphaFold 2, which won the CASP14 protein-structure assessment in November 2020 with a median score of 92.4 out of 100. MIT Technology Review cited him on the accuracy gap over rival teams. With Demis Hassabis he shared half of the 2024 Nobel Prize in Chemistry for protein structure prediction, the other half going to David Baker for computational protein design, and the two also won the 2023 Breakthrough Prize in Life Sciences and the Albert Lasker Award for Basic Medical Research for leading the project.\n\nhttps://en.wikipedia.org/wiki/AlphaFold\nhttps://www.technologyreview.com/2020/11/30/1012712/deepmind-protein-folding-ai-solved-biology-science-drugs-disease/\nPicture: John Jumper, 2024 Nobel Prize Laureate in Chemistry.jpg, Arthur Petron, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:John_Jumper,_2024_Nobel_Prize_Laureate_in_Chemistry.jpg", "size": 3.59, "position": { "x": -10.2, "y": 58.46, "z": 0 }, "images": ["f644f89e-34cd-4707-bc8f-0c2061bf49d3"], "nodeImage": "f644f89e-34cd-4707-bc8f-0c2061bf49d3" }, { "ID": "ai-p-john-mccarthy", "name": "John McCarthy", "type": "Person", "properties": { "Country": "United States", "Role": "Coined the term artificial intelligence; Lisp; professor at Stanford", "Impact score": 58 }, "description": "John McCarthy (1927 to 2011) was an American computer scientist and a founder of artificial intelligence. As an assistant professor at Dartmouth from 1955 he co-wrote the proposal for the 1956 Dartmouth workshop and chose the name 'artificial intelligence'. He moved to MIT as a research fellow in 1956, developed the Lisp programming language in the late 1950s (published 1960), and from 1962 was a full professor at Stanford until 2000, where he helped build the Stanford AI laboratory. He received the Turing Award in 1971 for his contributions to AI.\n\nhttps://en.wikipedia.org/wiki/John_McCarthy_(computer_scientist)\nhttps://en.wikipedia.org/wiki/Dartmouth_workshop\nPicture: John McCarthy Stanford.jpg, null0, CC BY-SA 2.0, https://commons.wikimedia.org/wiki/File:John_McCarthy_Stanford.jpg", "size": 3.54, "position": { "x": -277.196, "y": 67.69, "z": 0 }, "images": ["c170f674-1966-426c-89ba-ae2ab3f12836"], "nodeImage": "c170f674-1966-426c-89ba-ae2ab3f12836" }, { "ID": "ai-p-joseph-weizenbaum", "name": "Joseph Weizenbaum", "type": "Person", "properties": { "Country": "United States", "Role": "Creator of ELIZA; MIT professor and critic of AI", "Impact score": 33 }, "description": "Joseph Weizenbaum (1923 to 2008) was a German-American computer scientist, born in Berlin, whose Jewish family fled Nazi Germany in 1936. He joined MIT as an associate professor in 1963 and became a full professor in 1970. His chatbot ELIZA (1966) showed how easily people read understanding into a program, and its reception alarmed him. His 1976 book Computer Power and Human Reason made him a prominent critic of AI and of the idea that human judgment can be replaced by computation. He died in Germany in 2008.\n\nhttps://en.wikipedia.org/wiki/Joseph_Weizenbaum", "size": 2.65, "position": { "x": -251.455, "y": 58.46, "z": 0 }, "images": ["f6359ab4-10ca-44dc-b0af-4b591272cac1"], "nodeImage": "f6359ab4-10ca-44dc-b0af-4b591272cac1" }, { "ID": "ai-p-juergen-schmidhuber", "name": "Jürgen Schmidhuber", "type": "Person", "properties": { "Country": "Switzerland", "Role": "Co-author of LSTM; scientific director of IDSIA", "Impact score": 48 }, "description": "Jürgen Schmidhuber (born 1963 in Munich) is a German computer scientist specialising in artificial neural networks. He has been scientific director of the Dalle Molle Institute for AI Research (IDSIA) in Switzerland since 1995, a professor at the Università della Svizzera italiana in Lugano from 2009 to 2021, and since 2021 director of the AI Initiative at KAUST in Saudi Arabia. With Sepp Hochreiter he published the 1997 LSTM paper. He has argued that Hinton, Bengio and LeCun failed to credit earlier pioneers when they shared the 2018 Turing Award, and Yann LeCun replied to the New York Times that he was 'manically obsessed with recognition'.\n\nhttps://en.wikipedia.org/wiki/J%C3%BCrgen_Schmidhuber\nPicture: Jürgen Schmidhuber, 2017 (cropped).jpg, ITU/R.Farrell, CC BY 2.0, https://commons.wikimedia.org/wiki/File:J%C3%BCrgen_Schmidhuber,_2017_(cropped).jpg", "size": 3.23, "position": { "x": -193.889, "y": 67.69, "z": 0 }, "images": ["a41407b0-d054-48c1-82ea-bc376682af69"], "nodeImage": "a41407b0-d054-48c1-82ea-bc376682af69" }, { "ID": "ai-hist-p-kaiming-he", "name": "Kaiming He", "type": "Person", "properties": { "Country": "United States", "Role": "First author of ResNet", "Impact score": 48 }, "description": "Kaiming He is a Chinese-American computer scientist working on computer vision and deep learning. He earned a bachelor's degree at Tsinghua University in 2007 and a PhD at the Chinese University of Hong Kong in 2011, worked at Microsoft Research Asia from 2011 to 2016 and at Facebook AI Research from 2016 to 2024, and became an associate professor at MIT in 2024, with a part-time role at Google DeepMind. He is one of the creators of ResNet, won CVPR best paper awards in 2009 and 2016 and the 2017 ICCV Marr Prize, and received the 2023 Future Science Prize.\n\nhttps://en.wikipedia.org/wiki/Kaiming_He\nhttps://arxiv.org/abs/1512.03385", "size": 3.23, "position": { "x": -95.704, "y": 58.46, "z": 0 } }, { "ID": "ai-p-lee-sedol", "name": "Lee Sedol", "type": "Person", "properties": { "Country": "South Korea", "Role": "Go professional; opponent of AlphaGo in 2016", "Impact score": 58 }, "description": "Lee Sedol (born 1983 in Sinan County, South Jeolla) is a South Korean former professional Go player who turned professional in 1995, reached 9 dan in 2003 and won 18 international titles, second only to Lee Chang-ho's 21. In March 2016 he lost 1 to 4 to Google DeepMind's AlphaGo in Seoul, winning Game 4. He called the result 'Lee Se-dol's defeat', not a defeat of mankind. In November 2019 he announced his retirement, saying: 'Even if I become the number one, there is an entity that cannot be defeated.'\n\nhttps://en.wikipedia.org/wiki/Lee_Sedol\nhttps://en.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol\nPicture: Lee Se-Dol - 2016 (cropped).jpg, LG Electronics, CC BY 2.0, https://commons.wikimedia.org/wiki/File:Lee_Se-Dol_-_2016_(cropped).jpg", "size": 3.54, "position": { "x": -81.474, "y": 58.46, "z": 0 }, "images": ["46a5895d-c074-4677-8f0d-b85102f4c56f"], "nodeImage": "46a5895d-c074-4677-8f0d-b85102f4c56f" }, { "ID": "ai-p-liang-wenfeng", "name": "Liang Wenfeng", "type": "Person", "properties": { "Country": "China", "Role": "Founder and CEO of DeepSeek; co-founder of High-Flyer", "Impact score": 48 }, "description": "Liang Wenfeng, born in 1985 in Wuchuan, Guangdong, is the founder and chief executive of DeepSeek and co-founder of the quantitative hedge fund High-Flyer, which he started in February 2016 with two classmates from Zhejiang University. DeepSeek was launched in May 2023 after High-Flyer announced that it would pursue artificial general intelligence. On 20 January 2025, the day DeepSeek released R1, he attended a symposium hosted by Premier Li Qiang, and on 17 February 2025 a private-enterprise symposium hosted by Xi Jinping. Wikipedia says he personally holds 1 percent of DeepSeek and the High-Flyer partnership holds 99 percent; Bloomberg reported in July 2026 that he had become the richest AI founder.\n\nhttps://en.wikipedia.org/wiki/Liang_Wenfeng\nhttps://apnews.com/article/deepseek-founder-liang-wenfeng-china-ai-0673d5c39d90108189cc31b88d85b9f8", "size": 3.23, "position": { "x": 175.375, "y": 58.46, "z": 0 } }, { "ID": "ai-p-mark-zuckerberg", "name": "Mark Zuckerberg", "type": "Person", "properties": { "Country": "United States", "Role": "Co-founder, chairman and CEO of Meta 2004-", "Impact score": 55 }, "description": "Mark Zuckerberg is co-founder, chairman and chief executive of Meta. He backed the open release of Meta's LLaMA models from February 2023 and of Llama 2, 3 and 4, a strategy that made Meta the main Western supplier of open-weight models until the line's reception cooled in 2025. In June 2025 he personally recruited researchers for a new 'superintelligence' team, with pay packages the New York Times reported at $1 million to $100 million, after Meta paid $14.3 billion for 49 percent of Scale AI and hired its chief executive Alexandr Wang. Yann LeCun, Meta's chief AI scientist, announced in November 2025 that he would leave.\n\nhttps://en.wikipedia.org/wiki/Meta_Superintelligence_Labs\nhttps://en.wikipedia.org/wiki/Llama_(language_model)\nPicture: Mark Zuckerberg 2019 (cropped).jpg, Anthony Quintano from Westminster, United States, CC BY 2.0, https://commons.wikimedia.org/wiki/File:Mark_Zuckerberg_2019_(cropped).jpg", "size": 3.45, "position": { "x": 63.023, "y": 58.46, "z": 0 }, "images": ["3b92d45e-da14-4d74-a771-bfb0abeee07c"], "nodeImage": "3b92d45e-da14-4d74-a771-bfb0abeee07c" }, { "ID": "ai-p-marvin-minsky", "name": "Marvin Minsky", "type": "Person", "properties": { "Country": "United States", "Role": "Co-founder of the MIT AI laboratory; co-author of Perceptrons", "Impact score": 60 }, "description": "Marvin Minsky (1927 to 2016) was an American mathematician and cognitive scientist who joined the MIT faculty in 1958 and spent his career there. In 1951 he built SNARC, a machine often considered the first artificial neural network; in 1959 he and John McCarthy began what became MIT's AI laboratory; he co-authored the Dartmouth proposal of 1955 and, with Seymour Papert, the 1969 book Perceptrons. He received the Turing Award in 1969 and wrote The Society of Mind (1986), which argues that intelligence emerges from the cooperation of many simple agents.\n\nhttps://en.wikipedia.org/wiki/Marvin_Minsky\nhttps://en.wikipedia.org/wiki/Perceptrons_(book)\nPicture: Marvin Minsky at OLPCb (3x4 cropped).jpg, The original uploader was Sethwoodworth at English Wikipedia., CC BY 3.0, https://commons.wikimedia.org/wiki/File:Marvin_Minsky_at_OLPCb_(3x4_cropped).jpg", "size": 3.59, "position": { "x": -258.295, "y": 58.46, "z": 0 }, "images": ["ccac552d-3bec-4afa-a1be-203619be2549"], "nodeImage": "ccac552d-3bec-4afa-a1be-203619be2549" }, { "ID": "ai-p-masayoshi-son", "name": "Masayoshi Son", "type": "Person", "properties": { "Country": "Japan", "Role": "Founder, chairman and CEO of SoftBank Group; chairman of Stargate", "Impact score": 45 }, "description": "Masayoshi Son is the founder, chairman and chief executive of SoftBank Group. He is chairman of the Stargate venture announced at the White House on 21 January 2025, and SoftBank led OpenAI's $40 billion round of spring 2025, approved a second instalment of $22.5 billion in October 2025 to complete a $30 billion commitment, and committed $30 billion to OpenAI's February 2026 round. Wikipedia's SoftBank article puts SoftBank's cumulative OpenAI investment at about $34.6 billion by the end of 2025.\n\nhttps://en.wikipedia.org/wiki/SoftBank_Group\nhttps://en.wikipedia.org/wiki/Stargate_LLC\nPicture: Masayoshi Son (P066533-522034, cropped).jpg, European Communities, 2025 / EC - Audiovisual Service, CC BY 4.0, https://commons.wikimedia.org/wiki/File:Masayoshi_Son_(P066533-522034,_cropped).jpg", "size": 3.13, "position": { "x": 190.395, "y": 58.46, "z": 0 }, "images": ["db5f406d-2b48-41ff-acf1-35eb7451229b"], "nodeImage": "db5f406d-2b48-41ff-acf1-35eb7451229b" }, { "ID": "ai-p-mustafa-suleyman", "name": "Mustafa Suleyman", "type": "Person", "properties": { "Country": "United Kingdom", "Role": "Co-founder of DeepMind; chief executive of Microsoft AI from 2024", "Impact score": 45 }, "description": "Mustafa Suleyman (born 1984 in London) is a British AI entrepreneur who co-founded DeepMind in 2010 and became its chief product officer. After Google acquired DeepMind in 2014 he became head of applied AI; in 2019 he left DeepMind for a policy role at Google. In 2022 he co-founded Inflection AI, and in March 2024 Microsoft named him executive vice president and chief executive of Microsoft AI.\n\nhttps://en.wikipedia.org/wiki/Mustafa_Suleyman\nPicture: Mustafa Suleyman photo (cropped).jpg, Christopher Wilson, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Mustafa_Suleyman_photo_(cropped).jpg", "size": 3.13, "position": { "x": -119.484, "y": 58.46, "z": 0 }, "images": ["a15439b6-f3c8-4c30-a897-5b72e823f7b8"], "nodeImage": "a15439b6-f3c8-4c30-a897-5b72e823f7b8" }, { "ID": "ai-boom-p-pete-hegseth", "name": "Pete Hegseth", "type": "Person", "properties": { "Country": "United States", "Role": "US Secretary of Defense (Secretary of War) 2025-", "Impact score": 40 }, "description": "Pete Hegseth is the US Defense Secretary (styled Secretary of War since 2025). In February 2026 he demanded that Anthropic make Claude available to the military for all lawful purposes, set a deadline of 27 February, and after Anthropic refused to drop its limits on mass domestic surveillance and fully autonomous weapons, moved to designate the company a supply chain risk, a status normally reserved for adversary firms. The designation was notified on 5 March 2026; a federal judge later enjoined it and, in August 2026, ruled the blacklisting illegal, according to Wikipedia.\n\nhttps://fortune.com/2026/03/06/pentagon-officially-defines-anthropic-as-supply-chain-risk/\nhttps://en.wikipedia.org/wiki/Anthropic", "size": 2.95, "position": { "x": 272.83, "y": 58.46, "z": 0 }, "images": ["18b54705-c666-4ff0-bf77-830ede07c296"], "nodeImage": "18b54705-c666-4ff0-bf77-830ede07c296" }, { "ID": "ai-p-rishi-sunak", "name": "Rishi Sunak", "type": "Person", "properties": { "Country": "United Kingdom", "Role": "UK Prime Minister, host of the 2023 AI Safety Summit", "Impact score": 45 }, "description": "Rishi Sunak was UK Prime Minister when the United Kingdom hosted the AI Safety Summit at Bletchley Park on 1 and 2 November 2023. He opened the summit, which produced the Bletchley Declaration signed by 28 countries and the European Union, and the launch of the UK AI Safety Institute, and held a live conversation with Elon Musk on X on 2 November. The summit placed the UK at the centre of the frontier-AI safety agenda for a year.\n\nhttps://en.wikipedia.org/wiki/AI_Safety_Summit\nhttps://www.bbc.com/news/uk-england-beds-bucks-herts-67273099\nPicture: Portrait of Prime Minister Rishi Sunak (cropped).jpg, Prime Minister's Office, OGL 3, https://commons.wikimedia.org/wiki/File:Portrait_of_Prime_Minister_Rishi_Sunak_(cropped).jpg", "size": 3.13, "position": { "x": 112.157, "y": 58.46, "z": 0 }, "images": ["9d187775-0d2a-40fe-842f-e8b07c21984d"], "nodeImage": "9d187775-0d2a-40fe-842f-e8b07c21984d" }, { "ID": "ai-p-sam-altman", "name": "Sam Altman", "type": "Person", "properties": { "Country": "United States", "Role": "Co-founder and chief executive of OpenAI", "Impact score": 70 }, "description": "Sam Altman (born 1985 in Chicago) co-founded the location-sharing app Loopt in 2005 at age 19, which Green Dot Corporation acquired in 2012 for 43.4 million dollars, and was president of the start-up accelerator Y Combinator from 2014 to 2019. In December 2015 he became co-chair of OpenAI with Elon Musk when the non-profit was announced, and he has been its chief executive since 2019. OpenAI's board removed him on 17 November 2023 and he returned days later (21 November according to Wikipedia).\n\nhttps://en.wikipedia.org/wiki/Sam_Altman\nhttps://techcrunch.com/2015/12/11/non-profit-openai-launches-with-backing-from-elon-musk-and-sam-altman\nPicture: Sam Altman TechCrunch SF 2019 Day 2 Oct 3 (cropped).jpg, TechCrunch, CC BY 2.0, https://commons.wikimedia.org/wiki/File:Sam_Altman_TechCrunch_SF_2019_Day_2_Oct_3_(cropped).jpg", "size": 3.85, "position": { "x": 119.737, "y": 58.46, "z": 0 }, "images": ["9557e745-6061-4251-9c71-937f3f6ea4b4"], "nodeImage": "9557e745-6061-4251-9c71-937f3f6ea4b4" }, { "ID": "ai-p-satya-nadella", "name": "Satya Nadella", "type": "Person", "properties": { "Country": "United States", "Role": "CEO of Microsoft", "Impact score": 55 }, "description": "Satya Nadella is the chief executive of Microsoft and the executive who tied the company to OpenAI. In the 22 July 2019 announcement of Microsoft's $1 billion investment he said AI is one of the most transformative technologies of our time. Under his leadership Microsoft built the Azure supercomputer for OpenAI (announced 19 May 2020), licensed GPT-3 exclusively (22 September 2020), and in January 2023 announced a further multiyear investment of $10 billion. His strategy gave OpenAI the compute for GPT-3 and later models and gave Microsoft early access to them.\n\nhttps://news.microsoft.com/2019/07/22/openai-forms-exclusive-computing-partnership-with-microsoft-to-build-new-azure-ai-supercomputing-technologies/\nhttps://en.wikipedia.org/wiki/OpenAI\nPicture: MS-Exec-Nadella-Satya-2017-08-31-22 (cropped).jpg, Brian Smale and Microsoft, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:MS-Exec-Nadella-Satya-2017-08-31-22_(cropped).jpg", "size": 3.45, "position": { "x": -34.541, "y": 58.46, "z": 0 }, "images": ["2431be26-bae5-4751-b742-dda69a902121"], "nodeImage": "2431be26-bae5-4751-b742-dda69a902121" }, { "ID": "ai-hist-p-sebastian-thrun", "name": "Sebastian Thrun", "type": "Person", "properties": { "Country": "United States", "Role": "Led the Stanley team; founded Google's self-driving car project", "Impact score": 48 }, "description": "Sebastian Thrun (born 1967 in Solingen, West Germany) is a German-American computer scientist who worked at Carnegie Mellon before joining Stanford in 2003, where he re-established the Stanford AI Lab and led Stanley, the vehicle that won the 2005 DARPA Grand Challenge. At Google he co-founded Google X and started Project Chauffeur, the self-driving car project begun in January 2009 that became Waymo. He also co-founded the online-learning company Udacity and served as CEO of the electric aircraft company Kitty Hawk.\n\nhttps://en.wikipedia.org/wiki/Sebastian_Thrun\nhttps://en.wikipedia.org/wiki/Waymo\nPicture: Sebastian Thrun by Christopher Michel - 6.jpg, Christopher Michel, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Sebastian_Thrun_by_Christopher_Michel_-_6.jpg", "size": 3.23, "position": { "x": -178.17, "y": 58.46, "z": 0 }, "images": ["5d328c55-adb9-4d97-8370-6286d9dedb22"], "nodeImage": "5d328c55-adb9-4d97-8370-6286d9dedb22" }, { "ID": "ai-hist-p-sepp-hochreiter", "name": "Sepp Hochreiter", "type": "Person", "properties": { "Country": "International", "Role": "Inventor of LSTM; professor at Johannes Kepler University Linz", "Impact score": 40 }, "description": "Sepp Hochreiter (born 1967 in Mühldorf, Germany) is a German computer scientist who has led the Institute for Machine Learning at Johannes Kepler University Linz since 2018, after heading its Institute of Bioinformatics from 2006 to 2018. In his 1991 diploma thesis at the Technical University of Munich he developed long short-term memory to address vanishing and exploding gradients in recurrent networks, and the main LSTM publication with Jürgen Schmidhuber followed in 1997.\n\nhttps://en.wikipedia.org/wiki/Sepp_Hochreiter\nPicture: Sepp Hochreiter - Buch Wien 2025.JPG, C.Stadler/Bwag, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Sepp_Hochreiter_-_Buch_Wien_2025.JPG", "size": 2.95, "position": { "x": -184.95, "y": 58.46, "z": 0 }, "images": ["fff874e9-ff1f-4296-9106-58467c4b5eb2"], "nodeImage": "fff874e9-ff1f-4296-9106-58467c4b5eb2" }, { "ID": "ai-p-sundar-pichai", "name": "Sundar Pichai", "type": "Person", "properties": { "Country": "United States", "Role": "CEO of Google 2015-, CEO of Alphabet 2019-", "Impact score": 45 }, "description": "Sundar Pichai has been chief executive of Google since 2015 and of its parent Alphabet since 2019. He led the company through its response to ChatGPT: the December 2022 'code red' (which Wikipedia's Bard article says he later denied having issued), the February 2023 announcement of Bard, the April 2023 merger of Google Brain and DeepMind, and the Gemini launch in December 2023, which he announced together with Demis Hassabis. The Bard demo error and the fall in Alphabet's share price put him under pressure from employees, as CNBC reported in February 2023.\n\nhttps://en.wikipedia.org/wiki/Sundar_Pichai\nhttps://en.wikipedia.org/wiki/Bard_(chatbot)\nPicture: Sundar Pichai - 2023 (cropped).jpg, Photographer: Lukasz Kobus (European Commission), CC BY 4.0, https://commons.wikimedia.org/wiki/File:Sundar_Pichai_-_2023_(cropped).jpg", "size": 3.13, "position": { "x": 55.843, "y": 58.46, "z": 0 }, "images": ["add1d9d7-36ee-4fb9-8c2e-f2afb44f94b8"], "nodeImage": "add1d9d7-36ee-4fb9-8c2e-f2afb44f94b8" }, { "ID": "ai-scale-p-timnit-gebru", "name": "Timnit Gebru", "type": "Person", "properties": { "Country": "United States", "Role": "AI ethics researcher; co-lead of Google's ethical AI team until December 2020", "Impact score": 48 }, "description": "Timnit Gebru is an AI researcher who co-led Google's ethical AI team until December 2020 and co-wrote 'On the Dangers of Stochastic Parrots' with Emily Bender, Angelina McMillan-Major and Margaret Mitchell. Google's head of AI Jeff Dean said the paper did not meet its publication bar; Gebru said the company forced her out, while Google said it respected her decision to resign. About 2,700 Google employees and more than 4,300 academics and supporters signed a letter condemning her exit. The dispute made her a leading voice on the social costs of ever larger language models.\n\nhttps://en.wikipedia.org/wiki/Timnit_Gebru\nhttps://www.technologyreview.com/2020/12/04/1013294/google-ai-ethics-research-paper-forced-out-timnit-gebru/\nPicture: Timnit Gebru crop.jpg, TechCrunch, CC BY 2.0, https://commons.wikimedia.org/wiki/File:Timnit_Gebru_crop.jpg", "size": 3.23, "position": { "x": -0.308, "y": 58.46, "z": 0 }, "images": ["b116e561-73ec-4a31-9727-1a7f6f6ea8fe"], "nodeImage": "b116e561-73ec-4a31-9727-1a7f6f6ea8fe" }, { "ID": "ai-hist-p-tomas-mikolov", "name": "Tomáš Mikolov", "type": "Person", "properties": { "Country": "International", "Role": "Lead author of word2vec", "Impact score": 40 }, "description": "Tomáš Mikolov (born 1982) is a Czech computer scientist known for neural language models and word representations. He earned his PhD at Brno University of Technology in 2012, where his early work applied recurrent neural networks to language modelling and, in 2010, to speech recognition. He worked at Google Brain from 2012 to 2014 and at Facebook AI Research from 2014 to 2020, was lead author of the 2013 word2vec papers, and in 2020 joined the Czech Technical University in Prague. He has said he proposed an idea similar to sequence-to-sequence learning before the Google paper and was not credited.\n\nhttps://en.wikipedia.org/wiki/Tom%C3%A1%C5%A1_Mikolov\nhttps://en.wikipedia.org/wiki/Seq2seq\nPicture: Tomáš Mikolov (2020).jpg, Jindřich Nosek (NoJin), CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Tom%C3%A1%C5%A1_Mikolov_(2020).jpg", "size": 2.95, "position": { "x": -133.364, "y": 58.46, "z": 0 }, "images": ["eb4598ac-ef2f-47d1-9c46-edf7d1dadeae"], "nodeImage": "eb4598ac-ef2f-47d1-9c46-edf7d1dadeae" }, { "ID": "ai-p-yann-lecun", "name": "Yann LeCun", "type": "Person", "properties": { "Country": "United States", "Role": "Creator of LeNet; first director of Facebook AI Research; Turing Award 2018", "Impact score": 72 }, "description": "Yann LeCun (born 1960 near Paris) is a French-American computer scientist. After a PhD in 1987 and a year as a postdoctoral researcher under Geoffrey Hinton in Toronto, he worked at Bell Labs from 1988 to 1996, where he developed LeNet. He became a professor at New York University's Courant Institute in 2003, joined Facebook as chief AI scientist in 2013 and became the first director of its AI research lab that December, and left Meta in late 2025 to co-found AMI Labs. He shared the 2018 Turing Award with Hinton and Yoshua Bengio.\n\nhttps://en.wikipedia.org/wiki/Yann_LeCun\nPicture: Laura Chaubard & Yann Le Cun - 2024 (53814052697) (cropped).jpg, Jérémy Barande, CC BY-SA 2.0, https://commons.wikimedia.org/wiki/File:Laura_Chaubard_%26_Yann_Le_Cun_-_2024_(53814052697)_(cropped).jpg", "size": 3.9, "position": { "x": -200.18, "y": 58.46, "z": 0 }, "images": ["c7700112-f2ee-4fa7-a153-0a38b0c65fe8"], "nodeImage": "c7700112-f2ee-4fa7-a153-0a38b0c65fe8" }, { "ID": "ai-p-yoshua-bengio", "name": "Yoshua Bengio", "type": "Person", "properties": { "Country": "Canada", "Role": "Professor at Montréal; founder of Mila; Turing Award 2018", "Impact score": 80 }, "description": "Yoshua Bengio (born 1964 in Paris, a Canadian citizen) has been a professor at the Université de Montréal since 1993 and founded Mila, the Quebec AI Institute, where he was scientific director until 2025. He co-authored the 2003 paper introducing the neural probabilistic language model, which learned word embeddings, and the 2014 paper with Dzmitry Bahdanau and Kyunghyun Cho on attention for machine translation; he advised Ian Goodfellow during his PhD, in which the GAN was invented. He shared the 2018 Turing Award with Geoffrey Hinton and Yann LeCun.\n\nhttps://en.wikipedia.org/wiki/Yoshua_Bengio\nhttps://arxiv.org/abs/1409.0473\nPicture: ICLR 2025 - Yoshua Bengio 02.jpg, Xuthoria, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:ICLR_2025_-_Yoshua_Bengio_02.jpg", "size": 4.08, "position": { "x": -111.674, "y": 58.46, "z": 0 }, "images": ["9a61a1ac-8e22-428f-98ed-d5240dc662b7"], "nodeImage": "9a61a1ac-8e22-428f-98ed-d5240dc662b7" }, { "ID": "ai-scale-o-amd", "name": "AMD", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Chip company", "Ticker": "AMD", "Founded": 1969, "Impact score": 45 }, "description": "AMD (Advanced Micro Devices) is the US chip company that is Nvidia's main rival in AI accelerators. On 6 October 2025 it announced a multi-year agreement under which OpenAI will deploy 6 gigawatts of AMD GPUs, starting with 1 gigawatt of Instinct MI450 chips in the second half of 2026, and gave OpenAI a warrant for up to 160 million shares, about 10% of the company. Its shares jumped more than 30% at the open on the news. AMD's MI325X was one of the chips whose sale to China the US allowed on a case-by-case basis in January 2026.\n\nhttps://www.forbes.com.au/?p=159936\nhttps://www.channelnews.com.au/amd-challenges-nvidia-dominance-with-landmark-openai-chip-agreement/\nhttps://pillsburylaw.com/en/news-and-insights/trump-advanced-ai-semiconductors-actions.html\nhttps://en.wikipedia.org/wiki/AMD\nPicture: 2485 Augustine Drive headquarters in Santa Clara, California.jpg, Coolcaesar, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:2485_Augustine_Drive_headquarters_in_Santa_Clara,_California.jpg", "size": [4.46, 4.46], "position": { "x": 247.492, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["15d9df9b-eef4-4546-acd3-f3baa3c33f7c"] }, { "ID": "ai-o-amazon", "name": "Amazon", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Big tech company", "Ticker": "AMZN", "Founded": 1994, "Impact score": 62 }, "description": "Amazon is the US technology and retail company (Nasdaq: AMZN) founded in 1994 and headquartered in Seattle and Arlington, Virginia. It became Anthropic's largest strategic backer, investing $1.25 billion in 2023, $2.75 billion in March 2024, $4 billion in November 2024 and a further $5 billion in 2026, according to Wikipedia's Amazon article, and Anthropic trains on Amazon's Trainium chips. In February 2026 it committed $50 billion to OpenAI's $110 billion round, of which $15 billion was to be paid first, with 2 gigawatts of Trainium capacity and exclusive third-party cloud status for OpenAI's Frontier platform, as Bloomberg reported.\n\nhttps://en.wikipedia.org/wiki/Amazon_(company)\nhttps://www.advisorperspectives.com/articles/2026/02/28/openai-finalizes-110-billion-730-billion-value\nPicture: Amazon Tower I topped out, June 2015.jpg, SounderBruce, CC BY-SA 2.0, https://commons.wikimedia.org/wiki/File:Amazon_Tower_I_topped_out,_June_2015.jpg", "size": [5.04, 5.04], "position": { "x": -96.461, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["2fbd92a5-fa00-47d2-8582-d9e7b821e4bf"] }, { "ID": "ai-o-anthropic", "name": "Anthropic", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "AI lab", "Ticker": "", "Founded": 2021, "Impact score": 75 }, "description": "Anthropic is the AI safety company founded in January 2021 by former OpenAI staff, among them the siblings Dario Amodei, who had been OpenAI's vice president of research, and Daniela Amodei, with Jared Kaplan, Jack Clark, Chris Olah, Ben Mann, Sam McCandlish and Tom Brown. Wikipedia says the Amodeis left OpenAI over directional differences. It is a public benefit corporation. On 28 May 2021 it announced $124 million of funding led by Jaan Tallinn, with the stated aim of building steerable, interpretable and robust large-scale AI systems, and in April 2022 it announced $580 million including $500 million from FTX. It focused on research before releasing a product.\n\nhttps://en.wikipedia.org/wiki/Anthropic\nhttps://techcrunch.com/2021/05/28/anthropic-is-the-new-ai-research-outfit-from-openais-dario-amodei-and-it-has-124m-to-burn/", "size": [5.4, 5.4], "position": { "x": 197.55, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["e5aac5bd-18c1-4fe5-9b6e-bf13845cb7d2"] }, { "ID": "ai-o-apple", "name": "Apple", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Big tech company", "Ticker": "AAPL", "Founded": 1976, "Impact score": 48 }, "description": "Apple was founded on 1 April 1976 in Los Altos, California, by Steve Jobs, Steve Wozniak and Ronald Wayne, is headquartered at Apple Park in Cupertino and trades on Nasdaq as AAPL. In AI it is the company behind Siri: it bought the voice-assistant company Siri Inc., a spin-off from SRI's DARPA-funded CALO project, in April 2010 according to SRI's account, announced Siri with the iPhone 4S on 4 October 2011 and shipped it on 14 October. Apple later said it would use LSTM networks for QuickType on the iPhone and for Siri.\n\nhttps://en.wikipedia.org/wiki/Apple_Inc.\nhttps://en.wikipedia.org/wiki/Siri\nhttps://en.wikipedia.org/wiki/SRI_International", "size": [4.57, 4.57], "position": { "x": -156.138, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["42ff790f-a8d4-491e-8f43-57addb7eb5c9"] }, { "ID": "ai-hist-o-carnegie-mellon", "name": "CMU", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "University", "Ticker": "", "Founded": 1900, "Impact score": 30 }, "description": "Carnegie Mellon University in Pittsburgh was founded in 1900 as the Carnegie Technical Schools. Allen Newell and Herbert Simon built one of the first AI laboratories there after creating the Logic Theorist; John McDermott of Carnegie Mellon wrote XCON for Digital Equipment Corporation; DARPA's 1974 cancellation of a speech-understanding contract of about three million dollars a year hit its researchers; and its Robotics Institute fielded the robots Sandstorm and H1ghlander, which finished second and third in the 2005 DARPA Grand Challenge.\n\nhttps://en.wikipedia.org/wiki/Carnegie_Mellon_University\nhttps://en.wikipedia.org/wiki/Xcon\nhttps://en.wikipedia.org/wiki/AI_winter", "size": [3.74, 3.74], "position": { "x": -233.626, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["a6ed30b0-174a-4a24-b1d2-7ac98204552f"] }, { "ID": "ai-o-center-for-ai-safety", "name": "CAIS", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Non-profit", "Ticker": "", "Founded": 2022, "Impact score": 48 }, "description": "The Center for AI Safety (CAIS) is a San Francisco non-profit focused on reducing societal-scale risks from AI. On 30 May 2023 it published the 'Statement on AI Risk', a single sentence calling mitigation of the risk of extinction from AI a global priority alongside pandemics and nuclear war, signed by Geoffrey Hinton, Yoshua Bengio, and the leaders of OpenAI, Google DeepMind and Anthropic. The statement and its press coverage made CAIS one of the best-known voices in the AI risk debate.\n\nhttps://www.safe.ai/statement-on-ai-risk\nhttps://en.wikipedia.org/wiki/Statement_on_AI_risk_of_extinction", "size": [4.57, 4.57], "position": { "x": 86.441, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["9dfcb0b7-37b3-426f-9180-4236e180b1f9"] }, { "ID": "ai-o-cornell-aero-lab", "name": "Cornell Aero", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "AI lab", "Ticker": "", "Founded": 1946, "Impact score": 26 }, "description": "The Cornell Aeronautical Laboratory in Buffalo, New York, was an applied research laboratory that Curtiss-Wright donated to Cornell University, to be run as a public trust. It operated under that name from 1946 to 1972, when Cornell reorganised it as the for-profit Calspan Corporation. Frank Rosenblatt developed the perceptron there in 1957, funded by the US Office of Naval Research, and the Mark I Perceptron was assembled at Cornell in 1959 and 1960.\n\nhttps://en.wikipedia.org/wiki/Cornell_Aeronautical_Laboratory\nhttps://en.wikipedia.org/wiki/Perceptron", "size": [3.46, 3.46], "position": { "x": -272.713, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-o-cac", "name": "China's CAC", "type": "Organisation", "properties": { "Country": "China", "Organisation kind": "Government body", "Ticker": "", "Founded": 2014, "Impact score": 40 }, "description": "The Cyberspace Administration of China (CAC) is the Chinese government's internet regulator. With six other national bodies it issued the Interim Measures for the Management of Generative Artificial Intelligence Services on 13 July 2023, in force from 15 August 2023, which require security assessments, algorithm filing, lawful training data and content in line with core socialist values. It is the main licensing and filing authority for Chinese chatbot services such as those of DeepSeek and Baidu.\n\nhttps://www.cnbc.com/2023/07/13/china-introduces-rules-governing-generative-ai-services-like-chatgpt.html\nhttps://en.wikipedia.org/wiki/Interim_Measures_for_the_Management_of_Generative_Artificial_Intelligence_Services", "size": [4.25, 4.25], "position": { "x": 95.026, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-o-darpa", "name": "DARPA", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Government body", "Ticker": "", "Founded": 1958, "Impact score": 72 }, "description": "The Defense Advanced Research Projects Agency, created on 7 February 1958 as ARPA after the Soviet launch of Sputnik 1, is the research agency of the US Department of Defense. Its Information Processing Techniques Office, led first by J. C. R. Licklider, funded much of American AI research: parts of Shakey the robot, the Strategic Computing Initiative of the 1980s, and later the CALO project behind Siri. DARPA also cancelled a speech-understanding contract of about three million dollars a year at Carnegie Mellon in 1974, an event often cited as a sign of the first AI winter, and it ran the 2005 Grand Challenge that Stanford's Stanley won.\n\nhttps://en.wikipedia.org/wiki/DARPA\nhttps://en.wikipedia.org/wiki/AI_winter\nhttps://en.wikipedia.org/wiki/DARPA_Grand_Challenge_(2005)\nPicture: Darpa-logo-2026.png, DARPA-PAO, CC BY 4.0, https://commons.wikimedia.org/wiki/File:Darpa-logo-2026.png", "size": [5.32, 5.32], "position": { "x": -256.917, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["25468b44-76ff-4d4f-bffb-88cdba5e9080"] }, { "ID": "ai-hist-o-dnnresearch", "name": "DNNresearch", "type": "Organisation", "properties": { "Country": "Canada", "Organisation kind": "AI startup", "Ticker": "", "Founded": 2012, "Impact score": 40 }, "description": "DNNresearch was a start-up founded in 2012 by Geoffrey Hinton and two of his graduate students at the University of Toronto, Alex Krizhevsky and Ilya Sutskever, to bring deep neural network expertise to industry. In March 2013 Google acquired it; Google confirmed that the three researchers would join the company but did not disclose financial terms, and Hinton said he would split his time between the University of Toronto and Google's Toronto office and Mountain View headquarters.\n\nhttps://www.informationweek.com/data-management/google-buys-machine-learning-startup\nhttps://www.infoworld.com/article/2267586/google-acquires-startup-focused-on-neural-networks.html", "size": [4.25, 4.25], "position": { "x": -137.812, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-o-deepseek", "name": "DeepSeek", "type": "Organisation", "properties": { "Country": "China", "Organisation kind": "AI lab", "Ticker": "", "Founded": 2023, "Impact score": 75 }, "description": "DeepSeek is a Chinese AI lab launched in May 2023 by Liang Wenfeng after his hedge fund High-Flyer announced that it would pursue artificial general intelligence; Wikipedia says Liang holds 1 percent and the High-Flyer partnership 99 percent. It released DeepSeek-V3 in December 2024, the open-weight reasoning model R1 on 20 January 2025, V3.1 on 21 August 2025 under an MIT licence, and a V4 family from 24 April 2026. The January 2025 release and free app triggered a record one-day fall in Nvidia's value, bans on government devices and an Italian block of the app. The FY2026 US defence authorisation act directs removal of DeepSeek-developed AI from Defense Department and intelligence-community devices, per a Congressional Research Service summary.\n\nhttps://en.wikipedia.org/wiki/DeepSeek\nhttps://en.wikipedia.org/wiki/Liang_Wenfeng", "size": [5.4, 5.4], "position": { "x": 175.375, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-o-european-commission", "name": "EU Commission", "type": "Organisation", "properties": { "Country": "International", "Organisation kind": "Government body", "Ticker": "", "Founded": 1958, "Impact score": 45 }, "description": "The European Commission is the executive body of the European Union and the author of its proposals for new laws. On 21 April 2021 it proposed the Artificial Intelligence Act, which Executive Vice-President Margrethe Vestager presented with the words that the higher the risk, the stricter the rule, and Internal Market Commissioner Thierry Breton with the words that AI is a means, not an end. The proposal paired a four-level risk framework with about 1 billion euros a year of AI investment and a European Artificial Intelligence Board. Parliament and Council reached a deal on 9 December 2023.\n\nhttps://www.euronews.com/my-europe/2021/04/21/the-higher-the-risk-the-stricter-the-rule-brussels-new-draft-rules-on-artificial-intellige\nhttps://en.wikipedia.org/wiki/Artificial_Intelligence_Act\nhttps://en.wikipedia.org/wiki/European_Commission\nPicture: Berlaymont building (DSCF7542-DSCF7548).jpg, Trougnouf (Benoit Brummer), CC BY 4.0, https://commons.wikimedia.org/wiki/File:Berlaymont_building_(DSCF7542-DSCF7548).jpg", "size": [4.46, 4.46], "position": { "x": 2.16, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["22fd7110-52ab-4b21-ae31-a7abb2c85e28"] }, { "ID": "ai-o-future-of-life", "name": "Future of Life", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Non-profit", "Ticker": "", "Founded": 2014, "Impact score": 50 }, "description": "The Future of Life Institute (FLI) is a non-profit that works on risks from advanced technology, co-founded by Max Tegmark. It published the March 2023 open letter 'Pause Giant AI Experiments', which called for a six-month pause on training systems more powerful than GPT-4 and was signed by more than 30,000 people according to Wikipedia. FLI hosted the letter on its own site and revisited it a year later in 'The Pause Letter: One year later'. It also maintains a timeline of EU AI Act developments used by journalists.\n\nhttps://futureoflife.org/open-letter/pause-giant-ai-experiments/\nhttps://futureoflife.org/ai/the-pause-letter-one-year-later/", "size": [4.64, 4.64], "position": { "x": -113.72, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-o-italy-garante", "name": "Garante", "type": "Organisation", "properties": { "Country": "Italy", "Organisation kind": "Court or regulator", "Ticker": "", "Founded": 1997, "Impact score": 43 }, "description": "The Garante per la protezione dei dati personali is Italy's independent data protection authority. It imposed a temporary ban on ChatGPT on 31 March 2023, citing the lack of a legal basis for training on personal data, an age check and a data breach, and lifted it on 28 April after OpenAI changed its disclosures. On 30 January 2025 it blocked the DeepSeek app in Italy and opened an investigation, ordering the companies behind it to stop processing Italian users' data. Both actions made it the first European regulator to restrict a major chatbot.\n\nhttps://www.privacylaws.com/news/italy-s-garante-imposes-and-then-withdraws-a-ban-on-chatgpt/\nhttps://www.wsls.com/tech/2025/01/30/deepseek-blocked-by-some-app-stores-in-italy-to-protect-users-personal-data/", "size": [4.38, 4.38], "position": { "x": 70.811, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["d7deab52-444d-4b67-b3f1-f3ab5808ad2c"] }, { "ID": "ai-o-google", "name": "Google", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Big tech company", "Ticker": "GOOGL", "Founded": 1998, "Impact score": 100 }, "description": "Google was founded on 4 September 1998 in Menlo Park, California, is headquartered at the Googleplex in Mountain View, and became a subsidiary of Alphabet Inc. in the restructuring announced on 10 August 2015; its shares trade on Nasdaq as GOOGL and GOOG. It became the main commercial home of deep learning in the 2010s: it started Google Brain in 2011 (Jeff Dean, Greg Corrado and Andrew Ng), bought Geoffrey Hinton's company DNNresearch in March 2013 and DeepMind in January 2014, released the TensorFlow library in November 2015, and published its neural machine translation system in 2016. Google Brain and DeepMind merged in April 2023.\n\nhttps://en.wikipedia.org/wiki/Google\nhttps://en.wikipedia.org/wiki/Google_Brain\nhttps://en.wikipedia.org/wiki/TensorFlow\nPicture: Googleplex HQ (cropped).jpg, Asoundd, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Googleplex_HQ_(cropped).jpg", "size": [6, 6], "position": { "x": -150.253, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["3ef81270-2110-4b7a-b01a-ae58b00adc1a"] }, { "ID": "ai-o-deepmind", "name": "DeepMind", "type": "Organisation", "properties": { "Country": "United Kingdom", "Organisation kind": "AI lab", "Ticker": "", "Founded": 2010, "Impact score": 100 }, "description": "Full name: Google DeepMind. DeepMind was founded in London in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman; Companies House lists its incorporation on 23 September 2010 and the official launch was on 15 November 2010. Early investors included Horizons Ventures and Founders Fund, with individuals such as Peter Thiel, Scott Banister and Elon Musk, and Jaan Tallinn was an early investor and adviser. Google confirmed its acquisition on 26 January 2014. It built the Deep Q-network and AlphaGo, which beat Lee Sedol in 2016. It is based at Platform 37 in King's Cross, and in April 2023 it merged with Google Brain to form Google DeepMind.\n\nhttps://en.wikipedia.org/wiki/Google_DeepMind\nhttps://find-and-update.company-information.service.gov.uk/company/07386350", "size": [6, 6], "position": { "x": -130.41, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["09ce060d-5e5e-4a3d-96f5-aafda1a68063"] }, { "ID": "ai-scale-o-huawei", "name": "Huawei", "type": "Organisation", "properties": { "Country": "China", "Organisation kind": "Big tech company", "Ticker": "", "Founded": 1987, "Impact score": 48 }, "description": "Huawei is the Chinese technology company whose HiSilicon unit designs the Ascend AI accelerators, the main domestic alternative to Nvidia after the US export controls. Its Ascend 910B is made by SMIC on a 7 nm-class process, and the 910C combines two of them. The US Bureau of Industry and Security said in May 2025 that using Huawei Ascend chips anywhere in the world violates US export controls.\n\nhttps://en.wikipedia.org/wiki/HiSilicon\nhttps://www.yahoo.com/news/trump-administration-rescinds-biden-ai-193352703.html\nPicture: Sidejo de Huawei en Ŝenĵeno 09.jpg, RG72, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Sidejo_de_Huawei_en_%C5%9Cen%C4%B5eno_09.jpg", "size": [4.57, 4.57], "position": { "x": 206.162, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["abdccd16-ccb1-404c-bafb-dc5a06fca31a"] }, { "ID": "ai-o-ibm", "name": "IBM", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Big tech company", "Ticker": "IBM", "Founded": 1911, "Impact score": 80 }, "description": "IBM was founded in 1911 as the Computing-Tabulating-Recording Company and renamed International Business Machines in 1924; it is headquartered in Armonk, New York, and trades on the New York Stock Exchange as IBM. In AI it is known for two public demonstrations: Deep Blue, which beat world chess champion Garry Kasparov in 1997, and Watson, which won the quiz show Jeopardy! against Ken Jennings and Brad Rutter in February 2011. IBM later struggled to turn Watson into a profitable business: according to The Atlantic (2023), it cost about four times what it brought in, and the Watson Health division, which took about 4 billion dollars to develop, was sold in 2022 for about 1 billion dollars.\n\nhttps://en.wikipedia.org/wiki/IBM\nhttps://en.wikipedia.org/wiki/Watson_(computer)\nhttps://www.theatlantic.com/technology/archive/2023/05/ibm-watson-irrelevance-chatgpt-generative-ai-race/673965/\nPicture: IBM CHQ - Oct 2014.jpg, Treesmittenex, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:IBM_CHQ_-_Oct_2014.jpg", "size": [5.53, 5.53], "position": { "x": -196.718, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["9fbe7d9a-5331-4e5b-b376-7cb4e1861fa8"] }, { "ID": "ai-o-mit", "name": "MIT", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "University", "Ticker": "", "Founded": 1861, "Impact score": 40 }, "description": "The Massachusetts Institute of Technology, founded in 1861 and based in Cambridge, Massachusetts, was home to much early AI work. Project MAC, launched in 1963 with ARPA funding, brought researchers together around time-sharing and artificial intelligence, and its groups became the nucleus of the AI Laboratory; in 2003 the AI Laboratory merged with the Laboratory for Computer Science to form CSAIL. Marvin Minsky, John McCarthy (research fellow from 1956), Claude Shannon and Joseph Weizenbaum worked there, and Weizenbaum's ELIZA was written at MIT.\n\nhttps://en.wikipedia.org/wiki/Massachusetts_Institute_of_Technology\nhttps://en.wikipedia.org/wiki/ELIZA", "size": [4.25, 4.25], "position": { "x": -251.532, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["815e2311-c4f1-44a7-a787-804db260cf5e"] }, { "ID": "ai-o-meta", "name": "Meta", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Big tech company", "Ticker": "META", "Founded": 2004, "Impact score": 50 }, "description": "Meta Platforms was founded on 4 February 2004 in Cambridge, Massachusetts, as TheFacebook, Inc., is headquartered in Menlo Park, California, and trades on Nasdaq as META; it took the name Meta in 2021, announced at Facebook Connect on 28 October. Its AI research group, Facebook AI Research (FAIR), was founded in 2013 and led by Yann LeCun until 2018; FAIR released fastText in 2016 and the PyTorch library in 2017, and the company later released the Llama language models, starting in February 2023.\n\nhttps://en.wikipedia.org/wiki/Meta_Platforms\nhttps://en.wikipedia.org/wiki/Meta_AI\nPicture: Meta HQ 2023.png, InvadingInvader, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Meta_HQ_2023.png", "size": [4.64, 4.64], "position": { "x": 60.864, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["6c02da57-821a-4e62-bf03-8c5ef028a2aa"] }, { "ID": "ai-o-microsoft", "name": "Microsoft", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Big tech company", "Ticker": "MSFT", "Founded": 1975, "Impact score": 90 }, "description": "Microsoft is the software and cloud company that became OpenAI's main backer. It invested $1 billion in OpenAI on 22 July 2019 and moved OpenAI's services to its Azure cloud, built a supercomputer in Azure exclusively for OpenAI that it announced on 19 May 2020, and on 22 September 2020 took an exclusive licence to GPT-3 while OpenAI kept the public API. Through its GitHub subsidiary it launched Copilot in 2021, and on 23 January 2023 it announced a further multiyear investment of $10 billion in OpenAI. MIT Technology Review argued in 2020 that the exclusive licence showed how AI's compute needs concentrate power in wealthy tech companies.\n\nhttps://news.microsoft.com/2019/07/22/openai-forms-exclusive-computing-partnership-with-microsoft-to-build-new-azure-ai-supercomputing-technologies/\nhttps://news.microsoft.com/source/features/innovation/openai-azure-supercomputer/\nhttps://www.technologyreview.com/2020/09/23/1008729/openai-is-giving-microsoft-exclusive-access-to-its-gpt-3-language-model/\nPicture: Building92microsoft.jpg, Coolcaesar, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Building92microsoft.jpg", "size": [5.77, 5.77], "position": { "x": -102.466, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["4fcbf0f4-6c86-48cd-a495-c8ba77697395"] }, { "ID": "ai-o-nvidia", "name": "Nvidia", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Chip company", "Ticker": "NVDA", "Founded": 1993, "Impact score": 100 }, "description": "Nvidia is the US chip company whose GPUs and CUDA software became the standard hardware for training and running large AI models. It started in graphics for games; AlexNet was trained on two of its GeForce GTX 580 cards, and its V100, A100, H100 and Blackwell data-centre chips followed. Demand from the generative-AI boom took its market value past $1 trillion intraday in May 2023, to $3 trillion in June 2024, $4 trillion in July 2025 and $5 trillion on 29 October 2025, and a record one-day loss of about $590 billion followed the DeepSeek-R1 release on 27 January 2025. It is also the main target of US export controls on AI chips to China, which it answered with the A800, H800 and H20 variants.\n\nhttps://en.wikipedia.org/wiki/Nvidia\nhttps://www.koreajoongangdaily.com/business/investors-dump-tech-stocks-after-record-593-billion-nvidia-loss/12473281\nPicture: 2788-2888 San Tomas Expwy.jpg, Coolcaesar, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:2788-2888_San_Tomas_Expwy.jpg", "size": [6, 6], "position": { "x": -174.813, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["11c3fc01-aca3-41ff-88f0-9677024498b5"] }, { "ID": "ai-o-openai", "name": "OpenAI", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "AI lab", "Ticker": "", "Founded": 2015, "Impact score": 100 }, "description": "OpenAI was announced on 11 December 2015 as a non-profit AI research company (the Wikipedia infobox gives 8 December as the founding date). Its founders were Elon Musk, Sam Altman, Ilya Sutskever, Greg Brockman, Trevor Blackwell, Vicki Cheung, Andrej Karpathy, Durk Kingma, John Schulman, Pamela Vagata and Wojciech Zaremba, with Altman and Musk as co-chairs. A total of 1 billion dollars was pledged by Musk, Altman, Brockman, Reid Hoffman, Jessica Livingston, Peter Thiel, Amazon Web Services and Infosys, although tax filings showed that only 133 million dollars had been received by 2021. In September 2016 the company moved into the Pioneer Building in San Francisco's Mission District.\n\nhttps://en.wikipedia.org/wiki/OpenAI\nhttps://techcrunch.com/2015/12/11/non-profit-openai-launches-with-backing-from-elon-musk-and-sam-altman\nPicture: 1515 Third Street.jpg, Coolcaesar, CC BY 4.0, https://commons.wikimedia.org/wiki/File:1515_Third_Street.jpg", "size": [6, 6], "position": { "x": -53.199, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["2eb7d2e1-959a-4d3b-a79c-85bc04f42ab5", "63037d6d-0ef5-4700-9255-1b14f217b8df"] }, { "ID": "ai-boom-o-oracle", "name": "Oracle", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Big tech company", "Ticker": "ORCL", "Founded": 1977, "Impact score": 50 }, "description": "Oracle is the US enterprise-software and cloud company (NYSE: ORCL) headquartered in Austin, Texas, whose chairman Larry Ellison owns 42.4 percent, per Wikipedia. It is a partner in the Stargate venture, disclosed four multi-billion-dollar contracts on 9 September 2025, and was reported to have signed a $300 billion, five-year computing deal with OpenAI; its shares rose about 36.7 percent on 10 September 2025 (Reuters). Wikipedia's article on the AI bubble records later stress: S&P's cut to one notch above junk in July 2026, a force majeure notice on a New Mexico data-centre project on 24 September 2026 with a 3 percent share fall, and a record five-year credit-default-swap spread of 261 basis points on 8 October 2026.\n\nhttps://en.wikipedia.org/wiki/Oracle_Corporation\nhttps://en.wikipedia.org/wiki/Artificial_intelligence_bubble", "size": [4.64, 4.64], "position": { "x": 182.164, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["fa3fb751-3355-4483-90cc-d2e3ab0c94cb"] }, { "ID": "ai-hist-o-sri", "name": "SRI", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "AI lab", "Ticker": "", "Founded": 1946, "Impact score": 43 }, "description": "SRI International was founded in 1946 as the Stanford Research Institute by trustees of Stanford University, separated from Stanford in 1970 and took its present name in 1977; it is based in Menlo Park, California. Its Artificial Intelligence Center built Shakey the robot from 1966 to 1972 with DARPA support, and the DARPA-funded CALO project, which SRI described as the largest AI project ever launched, produced the technology behind Siri. SRI launched Siri Inc. in 2007 and Apple acquired it in April 2010.\n\nhttps://en.wikipedia.org/wiki/SRI_International\nPicture: SRI International HQ.jpg, Coolcaesar at English Wikipedia, CC BY-SA 3.0, https://commons.wikimedia.org/wiki/File:SRI_International_HQ.jpg", "size": [4.38, 4.38], "position": { "x": -262.367, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["7864a1e2-5bd4-4cd0-8f2b-44dfce51eea9"] }, { "ID": "ai-o-softbank", "name": "SoftBank", "type": "Organisation", "properties": { "Country": "Japan", "Organisation kind": "Investor", "Ticker": "9984", "Founded": 1981, "Impact score": 62 }, "description": "SoftBank Group is the Tokyo-based investment holding company (Tokyo Stock Exchange: 9984) founded in 1981 by Masayoshi Son, headquartered at Tokyo Portcity Takeshiba. It led OpenAI's $40 billion round in spring 2025, committed a total of $30 billion and had invested about $34.6 billion by the end of 2025, and committed another $30 billion to OpenAI's $110 billion round of February 2026. It is a partner and 40 percent stakeholder in the Stargate venture and also bought Ampere Computing for $6.5 billion, completed in November 2025, according to Wikipedia's SoftBank article.\n\nhttps://en.wikipedia.org/wiki/SoftBank_Group\nhttps://www.advisorperspectives.com/articles/2026/02/28/openai-finalizes-110-billion-730-billion-value\nPicture: Portcity Takeshiba.jpg, Akonnchiroll, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Portcity_Takeshiba.jpg", "size": [5.04, 5.04], "position": { "x": -13.417, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["58ea58aa-dbeb-423c-8323-c6567543d672"] }, { "ID": "ai-o-stability-ai", "name": "Stability AI", "type": "Organisation", "properties": { "Country": "United Kingdom", "Organisation kind": "AI startup", "Ticker": "", "Founded": 2019, "Impact score": 62 }, "description": "Stability AI is the London-based company that released Stable Diffusion in August 2022. Wikipedia gives its founding as 2019 by Emad Mostaque and Cyrus Hodes, though a separate account of Mostaque's career puts it in late 2020, so the year is disputed. It supplied the compute for the open-weights image model developed by CompVis and Runway, and raised $101 million in October 2022 at a reported $1 billion valuation. Getty Images sued it in the English High Court in January 2023, a case Getty largely lost in November 2025, and artists sued it in the United States the same month. Mostaque resigned as CEO on 23 March 2024.\n\nhttps://en.wikipedia.org/wiki/Stability_AI\nhttps://en.wikipedia.org/wiki/Emad_Mostaque\nhttps://techcrunch.com/2022/10/17/stability-ai-the-startup-behind-stable-diffusion-raises-101m/", "size": [5.04, 5.04], "position": { "x": 32.566, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["760d145e-30d7-4726-ad23-88de835f9584"] }, { "ID": "ai-o-stanford", "name": "Stanford", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "University", "Ticker": "", "Founded": 1885, "Impact score": 60 }, "description": "Stanford University, founded in 1885 by Leland and Jane Stanford and located in Stanford, California, in Silicon Valley, has been a centre of AI research since John McCarthy became a full professor there in 1962. It was home to the MYCIN expert system, Ed Feigenbaum's Knowledge Systems Laboratory, the team that built Stanley, winner of the 2005 DARPA Grand Challenge under Sebastian Thrun, and Fei-Fei Li's ImageNet effort. SRI International began as the Stanford Research Institute.\n\nhttps://en.wikipedia.org/wiki/Stanford_University\nhttps://en.wikipedia.org/wiki/Mycin\nhttps://en.wikipedia.org/wiki/Stanley_(vehicle)", "size": [4.97, 4.97], "position": { "x": -246.322, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["d1818468-059b-4d66-822e-2fec420b1fc5"] }, { "ID": "ai-o-china-state-council", "name": "State Council", "type": "Organisation", "properties": { "Country": "China", "Organisation kind": "Government body", "Ticker": "", "Founded": 1954, "Impact score": 38 }, "description": "Full name: State Council of the People's Republic of China. The State Council is the chief administrative authority of the People's Republic of China. On 20 July 2017 it issued the New Generation Artificial Intelligence Development Plan, which set the goal of making China the world's primary AI innovation centre by 2030, with a core AI industry above 1 trillion yuan. The plan's own diagnosis was that China led in speech and visual recognition and ranked second in AI papers and patents, but lagged in basic theory, core algorithms, high-end chips and foundational software, a gap later US export controls widened.\n\nhttps://digichina.stanford.edu/?p=826\nhttps://digichina.stanford.edu/?p=863\nhttps://en.wikipedia.org/wiki/State_Council_of_China\nPicture: Zhongnanhai-south-gates-3439.jpg, User:Vmenkov, CC BY-SA 3.0, https://commons.wikimedia.org/wiki/File:Zhongnanhai-south-gates-3439.jpg", "size": [4.16, 4.16], "position": { "x": -66.48, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["d335e115-bf8b-43ff-aa8c-a7a2d6130bc1"] }, { "ID": "ai-o-tsmc", "name": "TSMC", "type": "Organisation", "properties": { "Country": "Taiwan", "Organisation kind": "Chip company", "Ticker": "TSM", "Founded": 1987, "Impact score": 50 }, "description": "TSMC, the Taiwan Semiconductor Manufacturing Company, is the foundry that makes the advanced chips designed by Nvidia, Apple, Broadcom and Qualcomm. It was founded on 21 February 1987 in Hsinchu, Taiwan. Nvidia's H100 is built on TSMC's 4N process, and a Blackwell design flaw reported in October 2024 was fixed together with TSMC. TSMC stopped shipping wafers to Huawei and HiSilicon in September 2020 under US rules, and in October 2024 told the Commerce Department of a potential export-control breach in which one of its advanced chips reached Huawei through another company; SemiAnalysis noted that Huawei's Ascend 910C depends on TSMC wafers. It trades as 2330 in Taipei and TSM in New York.\n\nhttps://en.wikipedia.org/wiki/TSMC\nhttps://en.wikipedia.org/wiki/Hopper_(microarchitecture)\nhttps://en.wikipedia.org/wiki/Blackwell_(microarchitecture)\nPicture: TSMC Global R＆D Center at night.jpg, 曾 成訓, CC BY 2.0, https://commons.wikimedia.org/wiki/File:TSMC_Global_R%EF%BC%86D_Center_at_night.jpg", "size": [4.64, 4.64], "position": { "x": 20.954, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["329c8c14-be69-423b-946a-09c7ae8f68f4"] }, { "ID": "ai-o-new-york-times", "name": "NYT", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Media company", "Ticker": "NYT", "Founded": 1851, "Impact score": 38 }, "description": "The New York Times Company publishes The New York Times. As a plaintiff it sued OpenAI and Microsoft in federal court in Manhattan on 27 December 2023, alleging that its articles were copied to train GPT models; in March 2025 Judge Sidney Stein let the core copyright claim proceed.\n\nhttps://www.cbsnews.com/amp/miami/news/lawsuit-against-openai-newspaper-copyright\nhttps://giga.law/daily-news/2025/3/27/judge-allows-media-organizations-copyright-suit-against-openai-to-proceed\nPicture: The New York Times Building at sunset, 2021-09-30.jpg, Jdforrester, CC BY 4.0, https://commons.wikimedia.org/wiki/File:The_New_York_Times_Building_at_sunset,_2021-09-30.jpg", "size": [4.16, 4.16], "position": { "x": 124.089, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["fe45c425-273c-4406-9835-98921c626a8f"] }, { "ID": "ai-o-white-house", "name": "White House", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Government body", "Ticker": "", "Founded": 1789, "Impact score": 55 }, "description": "The White House stands here for the US executive branch under successive administrations. It issued Executive Order 14110 on 30 October 2023 under President Biden, and under President Trump issued the revocation of that order on 20 January 2025, Executive Order 14179 on 23 January 2025, the AI Action Plan with three executive orders on 23 July 2025, and the ordering of federal agencies away from Anthropic in February 2026. Policy swung from safety-based oversight to export promotion and deregulation within a single year.\n\nhttps://en.wikipedia.org/wiki/Executive_Order_14110\nhttps://cset.georgetown.edu/article/trumps-plan-for-ai-recapping-the-white-houses-ai-action-plan/", "size": [4.82, 4.82], "position": { "x": -79.166, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["58adc7de-e6c0-4576-a575-6949580e6156"] }, { "ID": "ai-o-us-commerce-bis", "name": "BIS", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Government body", "Ticker": "", "Founded": 2002, "Impact score": 62 }, "description": "Full name: US Department of Commerce, Bureau of Industry and Security (BIS). The Bureau of Industry and Security (BIS) is the part of the US Commerce Department that regulates exports of sensitive and dual-use technology. It was formed on 18 April 2002 by renaming the Bureau of Export Administration, which had been established on 1 October 1987, and it maintains the Commerce Control List and the Entity List. BIS wrote the AI chip rules of 7 October 2022 and 17 October 2023, the AI diffusion rule of January 2025 and its rescission on 13 May 2025, and the January 2026 rule that allowed case-by-case licences for Nvidia H200 sales to China. Under Secretary Jeffrey Kessler told Congress in July 2026 that H200 shipments had been extremely rare.\n\nhttps://en.wikipedia.org/wiki/Bureau_of_Industry_and_Security\nhttps://en.fnnews.com/news/202607150755449382", "size": [5.04, 5.04], "position": { "x": 38.206, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["d5e6eebc-c0e4-4dba-88ec-2abeed943e5d"] }, { "ID": "ai-boom-o-pentagon", "name": "Pentagon", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Government body", "Ticker": "", "Founded": 1947, "Impact score": 38 }, "description": "The US Department of Defense, styled the Department of War since 2025, is the Pentagon-based department of the US military. In February and March 2026 it demanded that Anthropic let the military use Claude for all lawful purposes, and after Anthropic refused to drop its limits on mass domestic surveillance and fully autonomous weapons, formally designated the company a supply chain risk on 5 March 2026. Wikipedia's Anthropic article reports that BBC News said on 5 October 2026 that the department had stopped using Anthropic products, including its removal from Palantir's Maven Smart System. OpenAI announced a deal to replace Claude in classified military settings.\n\nhttps://fortune.com/2026/03/06/pentagon-officially-defines-anthropic-as-supply-chain-risk/\nhttps://en.wikipedia.org/wiki/Anthropic", "size": [4.16, 4.16], "position": { "x": 272.83, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["ef52a94c-d5de-4017-af22-7a86a88738af"] }, { "ID": "ai-o-university-of-toronto", "name": "U of Toronto", "type": "Organisation", "properties": { "Country": "Canada", "Organisation kind": "University", "Ticker": "", "Founded": 1827, "Impact score": 58 }, "description": "The University of Toronto was founded as King's College by royal charter on 15 March 1827 and took its present name in 1850; it is based around Queen's Park in Toronto, Ontario. It is closely associated with the development of deep learning: Geoffrey Hinton has been a professor there since 1987, his students Alex Krizhevsky and Ilya Sutskever built AlexNet there, and the company DNNresearch that they founded with Hinton was a university spin-out sold to Google in 2013.\n\nhttps://en.wikipedia.org/wiki/University_of_Toronto\nhttps://en.wikipedia.org/wiki/AlexNet\nhttps://www.infoworld.com/article/2267586/google-acquires-startup-focused-on-neural-networks.html\nPicture: Toronto University College 1.jpg, GoginkLobabi, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Toronto_University_College_1.jpg", "size": [4.91, 4.91], "position": { "x": -144.198, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["703809bd-740f-40f3-8d60-a8e7f977be32"] }, { "ID": "ai-o-wga", "name": "WGA", "type": "Organisation", "properties": { "Country": "United States", "Organisation kind": "Non-profit", "Ticker": "", "Founded": 1933, "Impact score": 48 }, "description": "The Writers Guild of America (WGA) is the US labour union for screenwriters in film, television and news, split into WGA West and WGA East. It called the strike of 2 May 2023, its first in fifteen years, with about 11,500 members according to Wikipedia, and negotiated the contract of September 2023 that restricts the use of writers' material to train AI models and the use of AI to reduce writers' pay or number. The strike made the guild the leading example of organised labour reacting to generative AI.\n\nhttps://www.npr.org/2023/05/02/1172876800/writers-guild-calls-strike\nhttps://en.wikipedia.org/wiki/2023_Writers_Guild_of_America_strike\nPicture: Writers Guild of America West Building.jpg, Amanda scott, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:Writers_Guild_of_America_West_Building.jpg", "size": [4.57, 4.57], "position": { "x": 79.368, "y": 49.76, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["57629334-e8d1-4e2f-81f7-9bab2450b58b"] }, { "ID": "ai-a-lighthill-report", "name": "Lighthill report", "type": "Reaction", "properties": { "Reaction kind": "Government report", "Stance": "Scepticism", "Date": "1973-01-01", "Impact score": 55 }, "description": "The Lighthill report, 'Artificial Intelligence: A General Survey', was written by the Cambridge mathematician Sir James Lighthill after the UK Science Research Council asked him in 1972 to review academic AI research. He completed it in July, the council discussed it in September and decided to publish it, and it appeared in 1973 in the volume 'Artificial Intelligence: A Paper Symposium'. It was highly critical of AI's failure to meet its 'grandiose objectives' and, at a debate at the Royal Institution on 9 May 1973 with Donald Michie, John McCarthy and Richard Gregory, found defenders as well as critics. The report formed the basis for the British government's decision to end support for AI research in most British universities and is the usual starting point of the first AI winter.\n\nhttps://en.wikipedia.org/wiki/Lighthill_report\nhttps://www.chilton-computing.org.uk/inf/literature/reports/lighthill_report/p001.htm\nhttps://en.wikipedia.org/wiki/AI_winter", "size": 4.32, "position": { "x": -245.72, "y": -16.14, "z": 0 } }, { "ID": "ai-hist-re-weizenbaum-critique", "name": "Weizenbaum's critique", "type": "Reaction", "properties": { "Reaction kind": "Expert warning", "Stance": "Opposition", "Date": "1976-01-01", "Impact score": 28 }, "description": "In 1976 Joseph Weizenbaum, the creator of ELIZA, published Computer Power and Human Reason, a book that turned him into one of the best-known critics of artificial intelligence. Shaken by how readily people confided in his simple program, he argued that human thought cannot be fully reduced to computation and that there is a difference between deciding, which can be programmed, and choosing, which needs human judgment and values. He concluded that computers should not replace people in social and ethical matters such as therapy or justice. The book was a rare warning from inside the field during the years of early optimism and shaped later debates about the proper role of machines.\n\nhttps://en.wikipedia.org/wiki/ELIZA\nhttps://en.wikipedia.org/wiki/Joseph_Weizenbaum", "size": 3.11, "position": { "x": -237.69, "y": -16.14, "z": 0 } }, { "ID": "ai-hist-re-fli-open-letter-2015", "name": "FLI open letter 2015", "type": "Reaction", "properties": { "Reaction kind": "Open letter", "Stance": "Alarm", "Date": "2015-01-12", "Impact score": 50 }, "description": "In January 2015 the Future of Life Institute published 'Research Priorities for Robust and Beneficial Artificial Intelligence: An Open Letter', drafted for the participants of its first conference in Puerto Rico over the first weekend of the year and made public on 12 January. It called for research on the societal impacts of AI and on preventing unsafe or uncontrollable systems, with a twelve-page companion document of research priorities. More than 150 people signed at the outset, including Stephen Hawking, Elon Musk, Google's research director Peter Norvig, Stuart Russell of Berkeley and the founders of DeepMind and Vicarious. It was the first widely noticed collective statement by AI researchers about the risks of their field and a forerunner of the 2023 pause letter and extinction statement.\n\nhttps://en.wikipedia.org/wiki/Open_Letter_on_Artificial_Intelligence\nhttps://futureoflife.org/ai-open-letter/", "size": 4.14, "position": { "x": -113.72, "y": -16.14, "z": 0 } }, { "ID": "ai-hist-re-alphago-shock-korea", "name": "AlphaGo shock in Korea", "type": "Reaction", "properties": { "Reaction kind": "Public reaction", "Stance": "Alarm", "Date": "2016-03-16", "Impact score": 60 }, "description": "In South Korea AlphaGo's victory over the national hero Lee Sedol in March 2016 was received as a 'shock' that mixed admiration with alarm. The JoongAng Ilbo editorial 'The AlphaGo shock' of 16 March argued that AI could soon outperform humans in games and then in other fields, that Korea's economy relied on hardware such as memory chips while undervaluing software, and that the country needed long-term planning, legal changes to speed up R&D investment and public education about the coming AI era. Lee Sedol said after Game 3 that his was 'Lee Se-dol's defeat', not a defeat of mankind. The government answered a day later with a 1 trillion won AI fund, and President Park Geun-hye called AI 'the fourth industrial revolution'.\n\nhttps://www.koreajoongangdaily.com/opinion/the-alphago-shock/11510655\nhttps://en.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol\nhttps://behindtheblack.com/?p=38271", "size": 4.47, "position": { "x": -86.808, "y": -16.14, "z": 0 } }, { "ID": "ai-hist-re-alphago-china", "name": "China reacts to AlphaGo", "type": "Reaction", "properties": { "Reaction kind": "Public reaction", "Stance": "Competitive response", "Date": "2016-03-18", "Impact score": 60 }, "description": "In China, where Go is a national pastime, AlphaGo's defeat of Lee Sedol set off a debate. Foreign Policy reported on 18 March 2016 that some Chinese professionals saw the match as a contest between Eastern wisdom and Western technology, while others, such as the Go expert Chen Zuyuan, said the East-West framing was overplayed; some users on the social network Weibo used the result to criticise Baidu's lag behind Google and China's restrictions on Google. Demis Hassabis said that more than 100 million people had followed the first game's live coverage, including 60 million in China, a figure the article presents as his claim. China's top-ranked player Ke Jie first said AlphaGo could not beat him and then said he might lose; in May 2017 authorities barred live coverage of his match against AlphaGo, which he lost.\n\nhttps://foreignpolicy.com/2016/03/18/china-go-chess-west-east-technology-artificial-intelligence-google/\nhttps://chinadigitaltimes.net/2017/05/minitrue-no-live-coverage-ke-jie-vs-alphago-games/", "size": 4.47, "position": { "x": -96.348, "y": -16.14, "z": 0 } }, { "ID": "ai-scale-re-alphazero-chess-debate", "name": "AlphaZero chess debate", "type": "Reaction", "properties": { "Reaction kind": "Public reaction", "Stance": "Scepticism", "Date": "2017-12-08", "Impact score": 28 }, "description": "After DeepMind's AlphaZero preprint of 5 December 2017 reported a 100-game match win over the Stockfish chess engine, grandmasters and engine developers weighed in. Garry Kasparov called it a remarkable achievement, even if it should have been expected after AlphaGo, while Hikaru Nakamura and Stockfish's developer Tord Romstad objected that Stockfish had been run with suboptimal settings, time controls and hardware. Chess.com collected reactions from top grandmasters and the Stockfish author on 8 December. DeepMind answered with a larger 1,000-game match in 2018 that AlphaZero won with 155 wins, 6 losses and 839 draws.\n\nhttps://www.chess.com/news/view/alphazero-reactions-from-top-gms-stockfish-author\nhttps://en.wikipedia.org/wiki/AlphaZero", "size": 3.11, "position": { "x": -60.547, "y": -16.14, "z": 0 } }, { "ID": "ai-scale-re-gpt2-too-dangerous", "name": "GPT-2 'too dangerous' row", "type": "Reaction", "properties": { "Reaction kind": "Expert warning", "Stance": "Scepticism", "Date": "2019-02-14", "Impact score": 35 }, "description": "When OpenAI announced GPT-2 on 14 February 2019 and declined to publish the full model, citing the risk of malicious use, press and researchers split. The Guardian headlined that the text generator 'may be too dangerous to release, say creators', and Vox's Kelsey Piper wrote that one of the coolest AI systems she had seen might also kick her out of a job. Others argued the threat was exaggerated: Anima Anandkumar of Caltech and Nvidia called the refusal the 'opposite of open', and on 19 February Hugh Zhang published an open letter in The Gradient asking OpenAI to release the model. In November 2019 OpenAI said it had seen no strong evidence of misuse and published the full model.\n\nhttps://en.wikipedia.org/wiki/GPT-2\nhttps://www.theverge.com/2019/2/21/18234500/ai-ethics-debate-researchers-harmful-programs-openai\nhttps://thegradient.pub/openai-please-open-source-your-language-model/", "size": 3.52, "position": { "x": -44.187, "y": -16.14, "z": 0 } }, { "ID": "ai-scale-re-gpt3-bloviator", "name": "GPT-3 'bloviator' critique", "type": "Reaction", "properties": { "Reaction kind": "Expert warning", "Stance": "Scepticism", "Date": "2020-08-22", "Impact score": 35 }, "description": "On 22 August 2020 MIT Technology Review published an opinion piece by the cognitive scientists Gary Marcus and Ernest Davis titled 'GPT-3, Bloviator: OpenAI's language generator has no idea what it's talking about'. They argued that GPT-3 produces fluent text without understanding, and that its comprehension is often seriously off. The article was one of the sharpest expert responses to the model's launch and set out the sceptical case against scaling alone, which Marcus pressed again after ChatGPT. Others took the opposite view: David Chalmers called GPT-3 one of the most interesting and important AI systems ever produced.\n\nhttps://www.technologyreview.com/2020/08/22/1007539/gpt3-openai-language-generator-artificial-intelligence-ai-opinion\nhttps://en.wikipedia.org/wiki/GPT-3", "size": 3.52, "position": { "x": -20.623, "y": -16.14, "z": 0 } }, { "ID": "ai-scale-re-gebru-exit", "name": "Gebru leaves Google", "type": "Reaction", "properties": { "Reaction kind": "Resignation", "Stance": "Opposition", "Date": "2020-12-03", "Impact score": 50 }, "description": "On 3 December 2020 Timnit Gebru, co-lead of Google's ethical AI team, said she had been forced out after Google objected to a paper she co-wrote on the risks of large language models. Google's head of AI Jeff Dean said the paper ignored too much relevant research and did not meet the bar for publication; Gebru said she did not resign and was fired, while Google said it respected her decision to resign. About 2,700 Google employees and more than 4,300 academics and supporters signed a letter condemning the departure, and nine members of Congress wrote to Google asking for an explanation. The case turned the ethics of building ever larger models into a workplace and public dispute.\n\nhttps://www.technologyreview.com/2020/12/04/1013294/google-ai-ethics-research-paper-forced-out-timnit-gebru/\nhttps://www.nytimes.com/2020/12/03/technology/google-researcher-timnit-gebru.html\nhttps://en.wikipedia.org/wiki/Timnit_Gebru", "size": 4.14, "position": { "x": -6.959, "y": -16.14, "z": 0 } }, { "ID": "ai-scale-re-eu-proposal-reactions", "name": "Reactions to EU AI draft", "type": "Reaction", "properties": { "Reaction kind": "Expert warning", "Stance": "Alarm", "Date": "2021-04-21", "Impact score": 35 }, "description": "When the European Commission published its draft AI rules on 21 April 2021, rights groups, lawmakers and industry criticised them from opposite sides. Sarah Chander of European Digital Rights said the draft gave companies too wide a scope for self-regulation. Green MEP Patrick Breyer said it would bring high-risk public facial recognition to the whole EU and undermine freedoms. Cecilia Bonefeld-Dahl of DIGITAL EUROPE warned that the rules could place an excessive burden on many providers. Euronews reported that biometric surveillance was the most contested point, with rights groups calling the exceptions worrying and industry warning that compliance costs could deter start-ups.\n\nhttps://www.euronews.com/my-europe/2021/04/21/the-higher-the-risk-the-stricter-the-rule-brussels-new-draft-rules-on-artificial-intellige", "size": 3.52, "position": { "x": 2.16, "y": -16.14, "z": 0 } }, { "ID": "ai-scale-re-lemoine-sentience", "name": "LaMDA sentience claim", "type": "Reaction", "properties": { "Reaction kind": "Expert warning", "Stance": "Alarm", "Date": "2022-06-11", "Impact score": 45 }, "description": "On 11 June 2022 The Washington Post reported that Blake Lemoine, a Google engineer, had told executives Blaise Aguera y Arcas and Jen Gennai that LaMDA had become sentient, citing its answers about self-identity, moral values and religion; Google placed him on paid administrative leave. Most scientists rejected the claim. Gary Marcus, David Pfau of DeepMind and Erik Brynjolfsson of Stanford were among the critics, Yann LeCun said such neural networks were not powerful enough to attain true intelligence, and Timnit Gebru called Lemoine a victim of a hype cycle driven by researchers and the media. The episode made 'AI sentience' a mainstream news topic five months before ChatGPT.\n\nhttps://www.washingtonpost.com/technology/2022/06/11/google-ai-lamda-blake-lemoine/\nhttps://www.theguardian.com/technology/2022/jun/12/google-engineer-ai-bot-sentient-blake-lemoine\nhttps://en.wikipedia.org/wiki/LaMDA", "size": 3.96, "position": { "x": 10.24, "y": -16.14, "z": 0 } }, { "ID": "ai-scale-re-artists-colorado", "name": "Artists protest AI prize", "type": "Reaction", "properties": { "Reaction kind": "Public reaction", "Stance": "Opposition", "Date": "2022-08-31", "Impact score": 45 }, "description": "After Jason Allen's Midjourney image won the Colorado State Fair on 29 August 2022 and he shared the result online, artists reacted with anger. Vice reported on 31 August that artists were pissed, and The New York Times followed on 2 September under the headline 'An A.I.-Generated Picture Won an Art Prize. Artists Aren't Happy.' CNN and The Washington Post covered the same dispute over whether prompting a model is art. Allen said he had broken no rules, and the fair's judges said they would have awarded him the prize anyway. The episode became the public symbol of the clash between artists and generative-AI tools that led to lawsuits in 2023.\n\nhttps://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html\nhttps://www.vice.com/en/article/an-ai-generated-artwork-won-first-place-at-a-state-fair-fine-arts-competition-and-artists-are-pissed/\nhttps://en.wikipedia.org/wiki/Th%C3%A9%C3%A2tre_D%27op%C3%A9ra_Spatial", "size": 3.96, "position": { "x": 33.813, "y": -16.14, "z": 0 } }, { "ID": "ai-scale-re-copilot-class-action", "name": "Copilot class action", "type": "Reaction", "properties": { "Reaction kind": "Lawsuit", "Stance": "Legal challenge", "Date": "2022-11-03", "Impact score": 55 }, "description": "In November 2022 programmers filed a class action in the United States against GitHub, Microsoft and OpenAI over Copilot. The complaint, dated 3 November 2022 and prepared by the Joseph Saveri Law Firm with the programmer-lawyer Matthew Butterick, alleged that Copilot reproduces open-source code without honouring licence terms such as attribution, and raised claims including breach of contract with GitHub's users and breach of privacy under California's CCPA. The Verge called it the lawsuit that could rewrite the rules of AI copyright. In September 2026 the Ninth Circuit upheld the dismissal of the DMCA claims without resolving the open-source licence claims, and Reuters reported that OpenAI and Microsoft had fended off part of the suit. The allegations have not been proven.\n\nhttps://en.wikipedia.org/wiki/GitHub_Copilot\nhttps://www.theverge.com/2022/11/8/23446821/microsoft-openai-github-copilot-class-action-lawsuit-ai-copyright-violation-training-data\nhttps://www.reuters.com/legal/government/openai-microsoft-fend-off-part-software-developer-lawsuit-over-ai-training-2026-09-16/", "size": 4.32, "position": { "x": 19.12, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-google-code-red", "name": "Google code red", "type": "Reaction", "properties": { "Reaction kind": "Corporate response", "Stance": "Competitive response", "Date": "2022-12-21", "Impact score": 60 }, "description": "In December 2022, weeks after ChatGPT's launch, Google's management reportedly declared a 'code red' because a chatbot that answered questions directly threatened its search advertising business. The New York Times reported on 21 December 2022 that Google leaders including CEO Sundar Pichai had held meetings to redefine the company's AI efforts, reassign teams and speed up product reviews; Wikipedia's Bard article notes that Pichai later denied to the Times that he issued the alert. The alert led to the rushed announcement of the Bard chatbot in February 2023 and, eventually, to the merger of Google Brain and DeepMind in April 2023. The episode is the clearest example in the record of a corporate competitive response to a public product rather than to a research paper.\n\nhttps://www.nytimes.com/2022/12/21/technology/ai-chatgpt-google-search.html\nhttps://en.wikipedia.org/wiki/Bard_(chatbot)", "size": 4.47, "position": { "x": 52.925, "y": -16.14, "z": 0 } }, { "ID": "ai-a-pause-letter", "name": "Pause letter", "type": "Reaction", "properties": { "Reaction kind": "Open letter", "Stance": "Alarm", "Date": "2023-03-22", "Impact score": 75 }, "description": "The open letter 'Pause Giant AI Experiments', published by the Future of Life Institute in March 2023 about a week after GPT-4's release, called on all AI labs to pause for at least six months the training of systems more powerful than GPT-4. More than 30,000 people eventually signed, according to Wikipedia, among them Yoshua Bengio, Stuart Russell, Elon Musk, Steve Wozniak, Yuval Noah Harari and Emad Mostaque. Critics split: Eliezer Yudkowsky said it did not go far enough, Timnit Gebru and others said it distracted from present harms, Bill Gates said a pause would not solve the challenges, and Sam Altman said it lacked technical nuance and that OpenAI was not training GPT-5. No pause took place, but the letter is credited with raising government urgency before the Senate hearings, the executive order and the Bletchley summit.\n\nhttps://futureoflife.org/open-letter/pause-giant-ai-experiments/\nhttps://en.wikipedia.org/wiki/Pause_Giant_AI_Experiments:_An_Open_Letter", "size": 4.9, "position": { "x": 63.672, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-hinton-leaves-google", "name": "Hinton leaves Google", "type": "Reaction", "properties": { "Reaction kind": "Resignation", "Stance": "Alarm", "Date": "2023-05-01", "Impact score": 70 }, "description": "On 1 May 2023 Geoffrey Hinton, the Turing Award winner who had worked at Google since 2013, told the New York Times that he had resigned so that he could 'talk about the dangers of AI without considering how this impacts Google', and that part of him regretted his life's work. He pointed to chatbots' rapid progress, the risk of misinformation, and the prospect of systems more intelligent than people. MIT Technology Review and The Guardian reported the same day and the next. The resignation of one of deep learning's founders gave the existential-risk argument unusual authority and was followed within a month by the CAIS extinction statement, which Hinton signed; he shared the 2024 Nobel Prize in Physics.\n\nhttps://www.nytimes.com/2023/05/01/technology/ai-google-chatbot-engineer-quits-hinton.html\nhttps://www.technologyreview.com/2023/05/01/1072478/deep-learning-pioneer-geoffrey-hinton-quits-google/", "size": 4.77, "position": { "x": 73.942, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-wga-strike", "name": "Writers' strike 2023", "type": "Reaction", "properties": { "Reaction kind": "Strike or protest", "Stance": "Opposition", "Date": "2023-05-02", "Impact score": 70 }, "description": "The Writers Guild of America, representing about 11,500 screenwriters, went on strike on 2 May 2023 for the first time in fifteen years, the Hollywood Reporter and NPR reported. Among the issues, alongside streaming pay, was the studios' use of generative AI, which came to public attention after ChatGPT: writers wanted limits on AI-written scripts and on training models on their work. The strike ran 146 days to 27 September 2023. The agreement reached on 24 September bars the use of writers' material to train AI models, bars AI from reducing writers' number or pay, and bars digital recreations, according to Wikipedia's article on the strike; members ratified it on 9 October 2023 by 99 percent. The actors' union SAG-AFTRA struck from July to November 2023 over similar concerns.\n\nhttps://www.npr.org/2023/05/02/1172876800/writers-guild-calls-strike\nhttps://en.wikipedia.org/wiki/2023_Writers_Guild_of_America_strike\nPicture: We are not in severance.jpg, Fabebk, CC BY-SA 4.0, https://commons.wikimedia.org/wiki/File:We_are_not_in_severance.jpg", "size": 4.77, "position": { "x": 43.235, "y": -26.41, "z": 0 }, "images": ["09779411-1898-444f-a1ba-e52932bc9ca7"] }, { "ID": "ai-boom-re-cais-statement", "name": "Statement on AI Risk", "type": "Reaction", "properties": { "Reaction kind": "Expert warning", "Stance": "Alarm", "Date": "2023-05-30", "Impact score": 70 }, "description": "On 30 May 2023 the Center for AI Safety published a one-sentence statement: 'Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.' Signatories included Geoffrey Hinton and Yoshua Bengio, more than a hundred AI professors, and the chief executives or leaders of OpenAI (Sam Altman), Google DeepMind (Demis Hassabis) and Anthropic (Dario Amodei), as well as Bill Gates. It came a month after Hinton left Google and two weeks after Altman's Senate testimony. Critics, including The Atlantic and MIT Technology Review, argued that talk of extinction distracted from present harms; supporters said it made the concern respectable. The risk claim is the signatories' view, not an established fact.\n\nhttps://www.safe.ai/statement-on-ai-risk\nhttps://en.wikipedia.org/wiki/Statement_on_AI_risk_of_extinction", "size": 4.77, "position": { "x": 86.441, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-wga-deal", "name": "WGA deal with AI limits", "type": "Reaction", "properties": { "Reaction kind": "Court ruling or settlement", "Stance": "Opposition", "Date": "2023-09-24", "Impact score": 45 }, "description": "On 24 September 2023 the Writers Guild of America and the Alliance of Motion Picture and Television Producers reached a tentative agreement that ended the 146-day writers' strike; Variety called it a 'historic contract agreement'. The contract sets rules on artificial intelligence: studios may not require writers to use AI, AI-generated material is not treated as source material, and writers' work cannot be used to train AI models without limits, as summarised on Wikipedia's article on the strike. Leaders ended the strike at 12:01 a.m. on 27 September, and members ratified the deal on 9 October 2023 with 99 percent in favour. It is the first major labour-contract text in the United States to address generative AI.\n\nhttps://variety.com/2023/biz/news/writers-strike-ends-wga-amptp-deal-2-1235733452/\nhttps://en.wikipedia.org/wiki/2023_Writers_Guild_of_America_strike", "size": 3.96, "position": { "x": 104.139, "y": -16.14, "z": 0 } }, { "ID": "ai-a-nyt-v-openai", "name": "NYT v. OpenAI", "type": "Reaction", "properties": { "Reaction kind": "Lawsuit", "Stance": "Legal challenge", "Date": "2023-12-27", "Impact score": 65 }, "description": "On 27 December 2023 The New York Times Company sued OpenAI and Microsoft in federal court in Manhattan, alleging that millions of its articles were copied to train GPT models and that the chatbots can reproduce its journalism; the Times has said the defendants are effectively taking billions of dollars' worth of work by its journalists, which is the Times' characterisation and not a court finding. In March 2025 Judge Sidney Stein dismissed some claims but let the core copyright infringement claim proceed, possibly to a jury trial, as AP-syndicated reports said. The case is the leading US test of whether training on copyrighted news is fair use, and sits beside Bartz v. Anthropic, Getty v. Stability and suits by authors. The claims are allegations until ruled on.\n\nhttps://www.cbsnews.com/amp/miami/news/lawsuit-against-openai-newspaper-copyright\nhttps://giga.law/daily-news/2025/3/27/judge-allows-media-organizations-copyright-suit-against-openai-to-proceed", "size": 4.62, "position": { "x": 43.235, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-musk-sues-openai", "name": "Musk sues OpenAI", "type": "Reaction", "properties": { "Reaction kind": "Lawsuit", "Stance": "Legal challenge", "Date": "2024-02-29", "Impact score": 55 }, "description": "On 29 February 2024 Elon Musk, an OpenAI co-founder who left its board in 2018, sued OpenAI and Sam Altman, alleging they had abandoned the organisation's founding non-profit mission in favour of profit and a close relationship with Microsoft. Courthouse News and Reuters reported the filing. In November 2024 Musk sought a preliminary injunction against OpenAI's conversion to a for-profit structure; the judge allowed fraud and unjust-enrichment claims to proceed in May 2025, and OpenAI countersued in April 2025. The claims were allegations until the jury decided in OpenAI's favour on 18 May 2026.\n\nhttps://en.wikipedia.org/wiki/Musk_v._Altman\nhttps://www.reuters.com/legal/elon-musk-sues-openai-ceo-sam-altman-breach-contract-2024-03-01/", "size": 4.32, "position": { "x": 134.659, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-trump-deepseek-wakeup", "name": "DeepSeek 'wake-up call'", "type": "Reaction", "properties": { "Reaction kind": "Government report", "Stance": "Competitive response", "Date": "2025-01-27", "Impact score": 35 }, "description": "In the days after DeepSeek-R1 triggered a technology sell-off, President Donald Trump called the Chinese start-up's rise a 'wake-up call' for America's AI industry and also described it as a positive development, according to news reports summarised in Wikipedia's DeepSeek article. The reaction framed DeepSeek as a competitiveness shock and fed the debate on chip export controls and on the Stargate investment announced a week earlier. US agencies including NASA, the Navy and the Pentagon's Defense Information Systems Agency moved to block the DeepSeek app on their systems.\n\nhttps://en.wikipedia.org/wiki/DeepSeek_(AI)\nhttps://www.verdict.co.uk/italys-garante-blocks-deepseek-over-data-privacy-concerns/\nhttps://tomsguide.com/computing/online-security/deepseek-ai-banned-by-nasa-us-navy-and-more-over-privacy-concerns", "size": 3.52, "position": { "x": 183.534, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-bartz-settlement", "name": "Bartz v Anthropic", "type": "Reaction", "properties": { "Reaction kind": "Court ruling or settlement", "Stance": "Legal challenge", "Date": "2025-09-05", "Impact score": 60 }, "description": "On 5 September 2025 Anthropic agreed to pay at least $1.5 billion to settle Bartz v. Anthropic, a class action by authors who alleged it trained Claude on pirated copies of their books. On 23 June 2025 Judge William Alsup had ruled that training on lawfully acquired digital copies was fair use but that acquiring and keeping pirated copies was not and required a trial. NPR called the settlement the first of its kind in an AI copyright case. A judge approved it in July 2026, which Wikipedia's Anthropic article describes as the largest known settlement in a US copyright case. The settlement is a payment to resolve alleged infringement, not an admission, and it left the fair-use ruling for training as persuasive but non-binding precedent.\n\nhttps://www.npr.org/2025/09/05/nx-s1-5529404/anthropic-settlement-authors-copyright-ai\nhttps://en.wikipedia.org/wiki/Anthropic", "size": 4.47, "position": { "x": 236.816, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-getty-v-stability-ruling", "name": "Getty v Stability ruling", "type": "Reaction", "properties": { "Reaction kind": "Court ruling or settlement", "Stance": "Legal challenge", "Date": "2025-11-04", "Impact score": 35 }, "description": "On 4 November 2025 the High Court of England and Wales gave judgment in Getty Images v Stability AI, rejecting Getty's secondary copyright infringement claim over the Stable Diffusion image generator, according to analyses by the law firms Latham and Watkins and Mayer Brown. The case, filed in 2023, was the first major UK ruling on whether training an image model on copyrighted pictures infringes copyright. The ruling is a counterweight to US cases such as the New York Times suit and the Bartz settlement.\n\nhttps://www.lw.com/en/insights/getty-images-v-stability-ai-english-high-court-rejects-secondary-copyright-claim\nhttps://www.mayerbrown.com/en/insights/publications/2025/11/getty-images-v-stability-ai-what-the-high-courts-decision-means-for-rights-holders-and-ai-developers", "size": 3.52, "position": { "x": 253.553, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-openai-code-red", "name": "OpenAI code red", "type": "Reaction", "properties": { "Reaction kind": "Corporate response", "Stance": "Competitive response", "Date": "2025-12-02", "Impact score": 35 }, "description": "In early December 2025, about two weeks after Google released Gemini 3, OpenAI chief executive Sam Altman declared an internal 'code red' to marshal resources toward improving ChatGPT, according to The Information as relayed by AI Magazine and The Verge, which headlined it 'OpenAI declares code red as Google catches up in AI race' on 2 December 2025. Altman wrote that it was 'a critical time for ChatGPT' and OpenAI delayed other initiatives, including advertising plans. It mirrored Google's own code red of December 2022, three years earlier, and showed how the lead in consumer AI had moved. Accounts differ on how directly Gemini 3 caused the memo.\n\nhttps://aimagazine.com/news/openai-hits-code-red-after-googles-gemini-3\nhttps://en.wikipedia.org/wiki/Gemini_(language_model)", "size": 3.52, "position": { "x": 261.193, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-sora-shutdown", "name": "OpenAI drops Sora", "type": "Reaction", "properties": { "Reaction kind": "Corporate response", "Stance": "Competitive response", "Date": "2026-03-24", "Impact score": 45 }, "description": "On 24 March 2026 OpenAI announced it would discontinue Sora in its consumer app and API, saying 'we've decided to discontinue Sora' as it focused its compute, CBS News reported; the Wall Street Journal, which first reported it, said the video platform had waned in popularity, and Wikipedia says OpenAI gave no specific reason and that reports pointed to compute cost and a shift to enterprise products. The app closed on 26 April and the API on 24 September 2026; the Sora team continues research on world simulation for robotics. The decision ended a December 2025 licensing deal with Disney, whose spokesperson said the companies had a constructive collaboration. Deadline headlined it 'Disney's $1B Investment In OpenAI DOA As Sam Altman Pulls Sora Plug'. It is a rare case of a leading lab withdrawing a flagship generative product.\n\nhttps://www.cbsnews.com/news/sora-ai-openai-discontinues/\nhttps://en.wikipedia.org/wiki/Sora_(text-to-video_model)", "size": 3.96, "position": { "x": 276.839, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-musk-v-openai-verdict", "name": "Musk v. OpenAI verdict", "type": "Reaction", "properties": { "Reaction kind": "Court ruling or settlement", "Stance": "Legal challenge", "Date": "2026-05-18", "Impact score": 55 }, "description": "On 18 May 2026 a federal jury ruled for OpenAI and Sam Altman in Elon Musk's lawsuit over OpenAI's move from non-profit to for-profit, finding Musk's claims barred by the statute of limitations, as the Wall Street Journal reported. Jury selection began on 27 April, Musk testified on 28 to 30 April, Microsoft chief executive Satya Nadella testified on 11 May that Musk never raised concerns with him about Microsoft's investments, and closing arguments were on 15 May. In April 2026 Musk had amended the suit to ask that any damages go to OpenAI's charity and that Altman be removed from the board.\n\nhttps://en.wikipedia.org/wiki/Musk_v._Altman\nhttps://www.wsj.com/tech/ai/jury-sides-with-openai-sam-altman-in-case-brought-by-elon-musk-933240ff", "size": 4.32, "position": { "x": 286.463, "y": -16.14, "z": 0 } }, { "ID": "ai-boom-re-pacing-the-frontier", "name": "Pacing the Frontier", "type": "Reaction", "properties": { "Reaction kind": "Open letter", "Stance": "Alarm", "Date": "2026-07-28", "Impact score": 40 }, "description": "On 28 July 2026 more than 1,000 employees and executives of AI companies published an open letter titled 'Pacing the Frontier', asking the US government to back an international effort to build the technical and governance tools that would let the world deliberately slow automated AI development later, if needed. It does not call for a pause now. Coverage named signers including Anthropic chief executive Dario Amodei and co-founders Jack Clark and Jared Kaplan, OpenAI chief scientist Jakub Pachocki, Meta chief scientist Shengjia Zhao, Google's Anca Dragan and John Schulman of Thinking Machines; Anthropic and OpenAI were reported to back the initiative officially, while Sam Altman was described as more reserved. Reported signature counts ranged from about 1,100 to over 1,300 by early August. Critics, per Ground News, warned that such rules could weaken US competitiveness. It echoed the 2023 pause letter but came from insiders and asked for preparation rather than a halt.\n\nhttps://aiweekly.co/alerts/openai-anthropic-staff-urge-us-to-help-pace-frontier-ai\nhttps://www.trendingtopics.eu/1100-employees-at-openai-anthropic-meta-and-google-call-for-ai-slowdown/", "size": 3.75, "position": { "x": 303.133, "y": -16.14, "z": 0 } }, { "ID": "ai-hist-rg-us-ai-rd-strategic-plan", "name": "US AI R&D plan", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "National strategy", "Status": "Superseded", "Date": "2016-10-13", "Impact score": 35, "URL": "https://obamawhitehouse.archives.gov/sites/default/files/whitehouse_files/microsites/ostp/NSTC/national_ai_rd_strategic_plan.pdf" }, "description": "Full name: National AI Research and Development Strategic Plan. The National Artificial Intelligence Research and Development Strategic Plan was transmitted on 13 October 2016 by the co-chairs of the NITRD Subcommittee of the US National Science and Technology Council, following White House actions announced on 3 May 2016 to prepare for the benefits and risks of AI and a direction to create a national strategy for AI research. It sets a framework for identifying scientific needs and priorities for federally funded AI research, beyond near-term capabilities, in seven strategies: long-term investment, human-AI collaboration, ethical, legal and societal implications, safety and security, shared public datasets, standards and benchmarks, and the national AI R&D workforce. It does not set agendas for individual agencies, which continue to follow their own missions and budgets. The document is a US government work in the public domain.\n\nhttps://obamawhitehouse.archives.gov/sites/default/files/whitehouse_files/microsites/ostp/NSTC/national_ai_rd_strategic_plan.pdf", "size": [9.8, 2.18], "position": { "x": -79.166, "y": -36.18, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-china-ai-plan-2017", "name": "China AI plan 2017", "type": "Regulation", "properties": { "Jurisdiction": "China", "Regulation kind": "National strategy", "Status": "In force", "Date": "2017-07-20", "Impact score": 60, "URL": "https://digichina.stanford.edu/?p=826" }, "description": "China's State Council issued the New Generation Artificial Intelligence Development Plan on 20 July 2017, after completing it on 8 July. The plan set staged goals: by 2020 AI technology and applications in step with the global leaders and a core AI industry above 150 billion yuan; by 2025 major breakthroughs in basic theory and a core industry above 400 billion yuan; by 2030 world-leading AI theory, technology and applications, a core industry above 1 trillion yuan and China as the primary AI innovation centre. It acknowledged that China lagged in basic theory, core algorithms, high-end chips and foundational software. A 2020 academic analysis in AI and Society links the plan to AlphaGo's victories, and Kai-Fu Lee called AlphaGo's May 2017 match in China a Sputnik moment; the DigiChina authors described the plan as aspirational and not a moonshot.\n\nhttps://digichina.stanford.edu/?p=826\nhttps://digichina.stanford.edu/?p=863\nhttps://link.springer.com/10.1007/s00146-020-00992-2", "size": [9.8, 2.67], "position": { "x": -71.247, "y": -45.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-eu-ai-act-proposal", "name": "EU AI Act proposal", "type": "Regulation", "properties": { "Jurisdiction": "European Union", "Regulation kind": "Law", "Status": "Superseded", "Date": "2021-04-21", "Impact score": 75, "URL": "https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:52021PC0206" }, "description": "On 21 April 2021 the European Commission proposed the Artificial Intelligence Act, the first comprehensive attempt to regulate AI. It sorted systems into four risk levels: no new rules for minimal risk such as spam filters and games, transparency duties for limited risk such as chatbots, conformity assessment before and after market entry for high-risk uses such as facial recognition and CV screening, and a ban on unacceptable uses such as government social scoring. Fines for breaches would reach 30 million euros or 6% of global annual turnover, whichever is higher. Margrethe Vestager said the higher the risk, the stricter the rule, and the plan paired the rules with about 1 billion euros a year of AI investment. Parliament and Council reached a deal on 9 December 2023, and the text was superseded by the final Act.\n\nhttps://www.euronews.com/my-europe/2021/04/21/the-higher-the-risk-the-stricter-the-rule-brussels-new-draft-rules-on-artificial-intellige\nhttps://www.brookings.edu/blog/techtank/2021/05/04/machines-learn-that-brussels-writes-the-rules-the-eus-new-ai-regulation/\nhttps://en.wikipedia.org/wiki/Artificial_Intelligence_Act", "size": [9.8, 2.89], "position": { "x": 2.16, "y": -52.11, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-rg-chips-act", "name": "CHIPS Act", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Law", "Status": "In force", "Date": "2022-08-09", "Impact score": 50, "URL": "https://www.congress.gov/bill/117th-congress/house-bill/4346" }, "description": "President Joe Biden signed the CHIPS and Science Act on 9 August 2022. It authorises roughly $280 billion for domestic research and semiconductor manufacturing and appropriates $52.7 billion for chips, including $39 billion of subsidies for building fabs on US soil and a 25% investment tax credit for manufacturing equipment. Recipients face guardrails: a ten-year ban on producing chips more advanced than 28 nm in China and Russia, and a September 2023 Commerce rule limiting capacity expansion there. The law came two months before the export controls of October 2022 and was the industrial half of the same strategy, to build chips at home and restrict China's access to the most advanced ones.\n\nhttps://en.wikipedia.org/wiki/CHIPS_and_Science_Act\nhttps://www.congress.gov/bill/117th-congress/house-bill/4346\nhttps://www.govinfo.gov/link/plaw/117/public/167?link-type=html", "size": [8.2, 2.5], "position": { "x": 31.153, "y": -45.29, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["542c3572-16b1-4d8f-8da0-1a8a2768fc32"] }, { "ID": "ai-a-us-export-controls-2022", "name": "Chip curbs 2022", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Export control", "Status": "Partly in force", "Date": "2022-10-07", "Impact score": 80, "URL": "https://www.bis.doc.gov/index.php/policy-guidance/advanced-computing-and-semiconductor-manufacturing-items-controls-to-prc" }, "description": "Full name: US advanced computing export controls of October 2022. On 7 October 2022 the US Commerce Department's Bureau of Industry and Security announced rules that restricted China's access to advanced AI chips and the equipment to make them. They added advanced computing chips and products containing them to the Commerce Control List, required licences for supercomputer and chip-development end uses in China, extended US export rules to some foreign-made items, and set licence requirements for Chinese fabs making logic chips at 16/14 nm or below, DRAM at 18 nm half-pitch or below and NAND with 128 layers or more. CSIS called it the most expansive export-control action in decades; a Taiwanese supplier executive warned in Nikkei Asia of casualties and consequences. Nvidia answered with the A800 for China within a month. The Wikipedia account notes that Assistant Secretary Thea Kendler linked the rules to China's AI and military goals.\n\nhttps://www.bis.doc.gov/index.php/policy-guidance/advanced-computing-and-semiconductor-manufacturing-items-controls-to-prc\nhttps://en.wikipedia.org/wiki/United_States_New_Export_Controls_on_Advanced_Computing_and_Semiconductors_to_China\nhttps://www.aibusiness.com/verticals/nvidia-offers-new-ai-chip-to-china-to-comply-with-export-ban\nPicture: document, 2022.10.07 BIS Press Release Advanced Computing and Semiconductor Manufacturing Controls FINAL.pdf, U.S. Department of Commerce Bureau of Industry and Security, Public domain, https://commons.wikimedia.org/wiki/File:2022.10.07_BIS_Press_Release_Advanced_Computing_and_Semiconductor_Manufacturing_Controls_FINAL.pdf", "size": [9.8, 2.96], "position": { "x": 20.954, "y": -36.18, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["9a842088-45f9-43dc-8140-270f0fdc5848"] }, { "ID": "ai-boom-rg-italy-chatgpt-ban", "name": "Italy blocks ChatGPT", "type": "Regulation", "properties": { "Jurisdiction": "Italy", "Regulation kind": "Ban or restriction", "Status": "Revoked or rescinded", "Date": "2023-03-31", "Impact score": 60, "URL": "https://www.privacylaws.com/news/italy-s-garante-imposes-and-then-withdraws-a-ban-on-chatgpt/" }, "description": "Full name: Italy's Garante blocks ChatGPT. On 31 March 2023 Italy's data protection authority, the Garante, imposed a temporary ban on ChatGPT, the first Western regulator to do so. According to Privacy Laws and Business, the order followed a data breach affecting user conversations and subscriber payment information and cited inadequate notice to users and a weak legal basis for training on personal data; Bloomberg added the absence of an age check for children under 13. Digital Policy Alert dates the order to 30 March. OpenAI added privacy information, an opt-out form for training data and an age-verification tool, and ChatGPT was restored in Italy on 28 April 2023, as Bloomberg reported. The case became the template for European data-protection action against chatbots, including the Garante's block of DeepSeek in January 2025.\n\nhttps://www.privacylaws.com/news/italy-s-garante-imposes-and-then-withdraws-a-ban-on-chatgpt/\nhttps://www.bloomberg.com/news/articles/2023-04-28/openai-restores-chatgpt-in-italy-addresses-regulator-demands", "size": [9.8, 2.67], "position": { "x": 43.235, "y": -45.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-rg-china-genai-measures", "name": "China genAI rules", "type": "Regulation", "properties": { "Jurisdiction": "China", "Regulation kind": "Law", "Status": "In force", "Date": "2023-07-13", "Impact score": 55, "URL": "https://www.loc.gov/item/global-legal-monitor/2023-07-18/china-generative-ai-measures-finalized/" }, "description": "Full name: China's Interim Measures for generative AI services. China's Interim Measures for the Management of Generative Artificial Intelligence Services were finalised and published on 13 July 2023 by the Cyberspace Administration of China together with six other national bodies, and took effect on 15 August 2023. They apply to public-facing generative AI services in China and require adherence to law and to core socialist values, bar content that subverts state power or harms national security, require legitimately sourced training data, require labelling of generated images and video, and require security assessments and algorithm filing for services with public-opinion influence. CNBC called them China's first-of-their-kind rules for services like ChatGPT. They were an early national framework for generative AI, ahead of the EU AI Act's entry into force.\n\nhttps://www.cnbc.com/2023/07/13/china-introduces-rules-governing-generative-ai-services-like-chatgpt.html\nhttps://www.loc.gov/item/global-legal-monitor/2023-07-18/china-generative-ai-measures-finalized/", "size": [9.8, 2.59], "position": { "x": 43.235, "y": -36.18, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-rg-export-update-2023", "name": "Chip curbs 2023", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Export control", "Status": "Partly in force", "Date": "2023-10-17", "Impact score": 60, "URL": "https://www.bis.doc.gov/index.php/policy-guidance/advanced-computing-and-semiconductor-manufacturing-items-controls-to-prc" }, "description": "Full name: US update of the AI chip export controls, October 2023. On 17 October 2023 the US Bureau of Industry and Security released three rules that tightened the controls of October 2022: two interim final rules, on advanced computing and on semiconductor manufacturing equipment, and a final rule adding Chinese entities to the Entity List. The advanced-computing rule amended the controls on classifications 3A090 and 4A090, extended licence requirements to China and further country groups, and covered chips including Nvidia's A800 and H800, which had been designed to fit under the 2022 limits. CNBC headlined it as a ban on more AI chips including the H800. Law firm accounts differ on the Federal Register date (19 or 25 October 2023), and the rule took effect about 30 days after publication. Nvidia responded with the H20, L20 and L2.\n\nhttps://www.cnbc.com/2023/10/17/us-bans-export-of-more-ai-chips-including-nvidia-h800-to-china.html\nhttps://www.akingump.com/en/insights/alerts/commerce-imposes-significant-new-controls-on-advanced-semiconductors\nhttps://www.businessday.co.za/bd/companies/telecoms-and-technology/2023-11-24-nvidia-to-delay-launch-of-ai-chip-in-china-say-sources/", "size": [9.8, 2.67], "position": { "x": 65.44, "y": -52.11, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-eo-14110", "name": "EO 14110", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Executive order", "Status": "Revoked or rescinded", "Date": "2023-10-30", "Impact score": 65, "URL": "https://en.wikipedia.org/wiki/Executive_Order_14110" }, "description": "Full name: Executive Order 14110 on Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. Executive Order 14110, 'Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence', was signed by President Joe Biden on 30 October 2023. Wikipedia calls it the most comprehensive US governance action on AI up to that point. It told agencies to appoint chief AI officers, tasked the Department of Homeland Security with AI security guidelines and NIST with a generative-AI resource, called for watermarking of AI-generated content, and set federal procurement and use policy; it did not create a licensing regime for advanced models. Senator Ted Cruz and industry groups criticised it as a threat to innovation. It followed the 2023 pause letter and Senate hearings and came two days before the Bletchley summit. President Donald Trump revoked it within hours of taking office on 20 January 2025.\n\nhttps://en.wikipedia.org/wiki/Executive_Order_14110", "size": [9.8, 2.75], "position": { "x": 110.861, "y": -36.18, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["8801e21a-796a-4f71-8e67-1439273c2c81"] }, { "ID": "ai-boom-rg-bletchley-declaration", "name": "Bletchley Declaration", "type": "Regulation", "properties": { "Jurisdiction": "International", "Regulation kind": "International declaration", "Status": "Signed not binding", "Date": "2023-11-01", "Impact score": 55, "URL": "https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration/the-bletchley-declaration-by-countries-attending-the-ai-safety-summit-1-2-november-2023" }, "description": "The Bletchley Declaration was signed on 1 November 2023 at the AI Safety Summit by 28 countries and the European Union, including the United States, the United Kingdom and China. It states that frontier AI poses risks of 'shared concern', commits signatories to cooperate on identifying and managing them, and led to the commissioning of the International AI Safety Report. The Guardian reported it under the headline about the declaration of AI's 'catastrophic' danger. It is a political declaration, not a binding treaty, and was followed by the Seoul and Paris summits, at the second of which the United States and United Kingdom declined to sign the final statement.\n\nhttps://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration/the-bletchley-declaration-by-countries-attending-the-ai-safety-summit-1-2-november-2023\nhttps://www.theguardian.com/technology/2023/nov/01/uk-us-eu-and-china-sign-declaration-of-ais-catastrophic-danger\nPicture: UK Government hosts AI Summit at Bletchley Park (53302263013).jpg, UK Government, CC BY 2.0, https://commons.wikimedia.org/wiki/File:UK_Government_hosts_AI_Summit_at_Bletchley_Park_(53302263013).jpg", "size": [9.8, 2.59], "position": { "x": 112.157, "y": -52.11, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["75b004db-8cc1-4c11-8c17-e31b9bc3f181"] }, { "ID": "ai-a-eu-ai-act", "name": "EU AI Act", "type": "Regulation", "properties": { "Jurisdiction": "European Union", "Regulation kind": "Law", "Status": "Partly in force", "Date": "2024-08-01", "Impact score": 85, "URL": "https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689" }, "description": "The EU Artificial Intelligence Act, Regulation (EU) 2024/1689, is the first comprehensive horizontal law on AI. The European Parliament adopted the text on 13 March 2024 by 523 votes to 46 with 49 abstentions, the Council approved it on 21 May 2024, it was published in the Official Journal on 12 July 2024 and entered into force on 1 August 2024. It sorts systems by risk, bans a list of practices, sets duties for high-risk systems and for general-purpose AI models, and is enforced by the AI Office and national authorities. Obligations apply in stages: prohibitions and AI literacy from 2 February 2025, general-purpose AI duties from 2 August 2025, transparency duties from August 2026, and high-risk rules now deferred to December 2027 and August 2028 by the 2026 digital omnibus. It builds on the Commission's April 2021 proposal.\n\nhttps://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689\nhttps://commission.europa.eu/news/ai-act-enters-force-2024-08-01_en", "size": [8.2, 3.02], "position": { "x": 43.235, "y": -52.11, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-rg-diffusion-rule-2025", "name": "AI diffusion rule", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Export control", "Status": "Revoked or rescinded", "Date": "2025-01-15", "Impact score": 50, "URL": "https://www.akingump.com/en/insights/ai-law-and-regulation-tracker/bis-rescinds-ai-diffusion-rule-and-issues-new-guidance" }, "description": "Full name: US AI diffusion rule. The AI diffusion rule was issued by the Biden administration's Commerce Department in the last days of its term, published on 15 January 2025. It would have capped sales of advanced AI chips to most countries outside 18 allies and partners, with compliance due on 15 May 2025. Microsoft and Nvidia were among the companies that lobbied against it. The rule never took effect: Commerce rescinded it on 13 May 2025, two days before compliance began; White House AI adviser David Sacks had said the previous week that it alienated key allies and exceeded the government's export-control authority.\n\nhttps://www.akingump.com/en/insights/ai-law-and-regulation-tracker/bis-rescinds-ai-diffusion-rule-and-issues-new-guidance\nhttps://www.yahoo.com/news/trump-administration-rescinds-biden-ai-193352703.html", "size": [9.8, 2.5], "position": { "x": 178.277, "y": -45.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-a-eo-14110-revoked", "name": "EO 14110 revoked", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Executive order", "Status": "In force", "Date": "2025-01-20", "Impact score": 70, "URL": "https://en.wikipedia.org/wiki/Executive_Order_14110" }, "description": "Full name: Revocation of Executive Order 14110. On 20 January 2025, within hours of taking office, President Donald Trump rescinded Executive Order 14110, the Biden administration's October 2023 order on safe, secure and trustworthy AI, listing it among Biden-era actions he called 'unpopular, inflationary, illegal, and radical practices', according to Wikipedia's article on the order. The revocation removed the reporting and safety-testing duties the order had placed on developers and agencies. Three days later, on 23 January, Trump signed Executive Order 14179 on AI leadership, and on 23 July 2025 the White House published its AI Action Plan, setting a deregulatory course. The reversal is the clearest example in the record of US AI policy swinging between administrations.\n\nhttps://en.wikipedia.org/wiki/Executive_Order_14110\nhttps://en.wikipedia.org/wiki/2025_in_artificial_intelligence", "size": [9.8, 2.82], "position": { "x": 179.857, "y": -52.11, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-rg-eo-14179", "name": "EO 14179", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Executive order", "Status": "In force", "Date": "2025-01-23", "Impact score": 50, "URL": "https://www.wsgr.com/en/insights/white-house-releases-americas-ai-action-plan.html" }, "description": "Full name: Executive Order 14179, Removing Barriers to American Leadership in Artificial Intelligence. President Donald Trump signed Executive Order 14179 on 23 January 2025, three days after revoking Biden's Executive Order 14110. The order, which Wikipedia's 2025 timeline describes as revoking the Biden AI safety order, directed the administration to develop an action plan to sustain and enhance US AI dominance and to review actions taken under the earlier order. It led to the AI Action Plan of 23 July 2025. Law-firm summaries describe it as the start of the administration's deregulatory AI policy.\n\nhttps://www.wsgr.com/en/insights/white-house-releases-americas-ai-action-plan.html\nhttps://en.wikipedia.org/wiki/2025_in_artificial_intelligence\nPicture: document, Executive Order 14179.pdf, Donald Trump, Public domain, https://commons.wikimedia.org/wiki/File:Executive_Order_14179.pdf", "size": [9.8, 2.5], "position": { "x": 188.677, "y": -45.29, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["e7b90c5a-4250-4ea8-aae5-f59d30bd4d45"] }, { "ID": "ai-boom-rg-italy-deepseek-block", "name": "Italy blocks DeepSeek", "type": "Regulation", "properties": { "Jurisdiction": "Italy", "Regulation kind": "Ban or restriction", "Status": "In force", "Date": "2025-01-30", "Impact score": 28, "URL": "https://www.wsls.com/tech/2025/01/30/deepseek-blocked-by-some-app-stores-in-italy-to-protect-users-personal-data/" }, "description": "Full name: Italy's Garante blocks DeepSeek. On 30 January 2025 Italy's data protection authority, the Garante, ordered DeepSeek's operators to restrict processing of Italian users' data 'urgently and with immediate effect', blocking the app, and opened an investigation after calling DeepSeek's answers to its questions 'completely insufficient'. According to the Garante's statement as reported by AP, the companies said they do not operate in Italy and that European law does not apply to them, which the authority disputed. Reuters reported that the app was no longer available in Apple and Google app stores in Italy, and Ireland's Data Protection Commission asked DeepSeek for information. The action repeated the Garante's 2023 ban on ChatGPT and was followed by restrictions on government devices in the US and Australia.\n\nhttps://www.wsls.com/tech/2025/01/30/deepseek-blocked-by-some-app-stores-in-italy-to-protect-users-personal-data/\nhttps://mcdn.i-scmp.com/news/world/europe/article/3296854/italy-blocks-chinese-ai-app-deepseek-opens-investigation-data-use", "size": [9.8, 1.97], "position": { "x": 179.857, "y": -36.18, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-rg-ai-act-prohibitions", "name": "AI Act bans apply", "type": "Regulation", "properties": { "Jurisdiction": "European Union", "Regulation kind": "Law", "Status": "In force", "Date": "2025-02-02", "Impact score": 40, "URL": "https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai" }, "description": "Full name: EU AI Act: prohibitions and AI literacy apply. On 2 February 2025, six months after the EU Artificial Intelligence Act entered into force, its first obligations began to apply: the ban on eight categories of unacceptable-risk practices and the duty on providers and deployers to promote AI literacy, according to the European Commission's AI Act page. The date follows from the Act's six-month delay for prohibitions (Wikipedia derives it from the 1 August 2024 entry into force). It was the first binding stage of the Act, coming a week before the Paris AI Action Summit, where the United States and United Kingdom declined to sign the final statement. A ninth prohibition, on AI systems that generate non-consensual intimate imagery or child sexual abuse material, was added by the 2026 digital omnibus and applies from December 2026.\n\nhttps://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai\nhttps://en.wikipedia.org/wiki/Artificial_Intelligence_Act", "size": [9.8, 2.3], "position": { "x": 193.257, "y": -52.11, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-rg-h20-licence-2025", "name": "H20 licence rule", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Export control", "Status": "Superseded", "Date": "2025-04-09", "Impact score": 55, "URL": "https://citizen.digital/news/us-licenses-nvidia-to-export-chips-to-china-official-says-n367718" }, "description": "Full name: US requires licences for Nvidia H20 sales to China. From 9 April 2025 the US government required Nvidia to obtain a licence to export its H20 chip to China, including Hong Kong and Macau, and several other countries. Nvidia told the SEC that its first-quarter results would include charges of up to about $5.5 billion tied to H20 inventory, purchase commitments and related reserves; in May it said the actual charge was about $1 billion lower because some materials were reused. Reuters reported that the curbs would cut $8 billion from the July-quarter sales. Analysts treated the requirement as a de facto ban. It left Chinese cloud companies such as Tencent, ByteDance and Alibaba looking for domestic substitutes, among them Huawei's Ascend 910C.\n\nhttps://citizen.digital/news/us-licenses-nvidia-to-export-chips-to-china-official-says-n367718\nhttps://www.malaymail.com/news/money/2025/08/09/nvidia-scores-us-export-licence-win-clears-path-for-ai-chip-sales-to-china/187008\nhttps://techblog.comsoc.org/2025/04/21/u-s-export-controls-on-nvidia-h20-ai-chips-enables-huaweis-910c-gpu-to-be-favored-by-ai-tech-giants-in-china/", "size": [9.8, 2.59], "position": { "x": 121.261, "y": -36.18, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-rg-diffusion-rescission-2025", "name": "Diffusion rule ends", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Export control", "Status": "In force", "Date": "2025-05-13", "Impact score": 50, "URL": "https://media.bis.gov/sites/default/files/documents/05.07%20Recission%20of%20AI%20Diffusion%20Press%20Release.pdf" }, "description": "Full name: Rescission of the US AI diffusion rule. On 13 May 2025 the Commerce Department said it was rescinding the Biden-era AI diffusion rule two days before it was due to take effect, and Bureau of Industry and Security staff were told not to enforce it. Under Secretary Jeffrey Kessler said the Trump administration would pursue a bold, inclusive strategy to spread American AI technology to trusted foreign countries and rejected what he called the previous administration's ill-conceived and counterproductive AI policies. BIS said a replacement rule would follow, and issued guidance that using Huawei Ascend chips anywhere in the world violates US export controls and warning against diversion of US chips. The move followed lobbying by Microsoft and Nvidia and concern about China's AI progress after DeepSeek.\n\nhttps://www.yahoo.com/news/trump-administration-rescinds-biden-ai-193352703.html\nhttps://www.akingump.com/en/insights/ai-law-and-regulation-tracker/bis-rescinds-ai-diffusion-rule-and-issues-new-guidance\nhttps://media.bis.gov/sites/default/files/documents/05.07%20Recission%20of%20AI%20Diffusion%20Press%20Release.pdf", "size": [9.8, 2.5], "position": { "x": 206.162, "y": -45.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-rg-h20-reversal-2025", "name": "H20 ban reversed", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Export control", "Status": "Superseded", "Date": "2025-07-14", "Impact score": 45, "URL": "https://the-decoder.com/nvidia-can-resume-exports-of-its-h20-ai-chip-to-china-after-a-us-policy-reversal/" }, "description": "Full name: US reverses the H20 ban and begins issuing licences. On 14 July 2025 Nvidia said it expected to resume H20 sales to China after the US government gave assurances that licences would be granted, reversing the April requirement that had amounted to a ban. In China Jensen Huang said the US had approved Nvidia filing licence applications to start shipping H20s; The Decoder reported that Huang had met Donald Trump the week before and that Nvidia would also sell its RTX PRO GPUs in China. On 8 August 2025 a US official told Reuters that Commerce had started issuing licences to Nvidia to export H20 chips to China, without saying how many or for what value. Huang had called the controls a failure in May, arguing they mainly helped Huawei.\n\nhttps://the-decoder.com/nvidia-can-resume-exports-of-its-h20-ai-chip-to-china-after-a-us-policy-reversal/\nhttps://www.hl.co.uk/news/nvidia-is-gearing-up-to-sell-h20-chips-to-china-again\nhttps://citizen.digital/news/us-licenses-nvidia-to-export-chips-to-china-official-says-n367718", "size": [9.8, 2.41], "position": { "x": 219.635, "y": -52.11, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-rg-ai-action-plan", "name": "US AI Action Plan", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "National strategy", "Status": "In force", "Date": "2025-07-23", "Impact score": 60, "URL": "https://cset.georgetown.edu/article/trumps-plan-for-ai-recapping-the-white-houses-ai-action-plan/" }, "description": "America's AI Action Plan was published by the White House on 23 July 2025 as a 23-page document with three pillars: Accelerate AI Innovation (15 subsections), Build American AI Infrastructure (8) and Lead in International AI Diplomacy and Security (7), the Center for Security and Emerging Technology reported. President Trump signed three executive orders with it: Accelerating Federal Permitting of Data Center Infrastructure, Promoting the Export of the American AI Technology Stack, and Preventing Woke AI in the Federal Government. It followed Executive Order 14179 of 23 January 2025 and set a deregulatory, export-promotion course for US AI policy, in contrast with the EU AI Act.\n\nhttps://cset.georgetown.edu/article/trumps-plan-for-ai-recapping-the-white-houses-ai-action-plan/\nhttps://www.wsgr.com/en/insights/white-house-releases-americas-ai-action-plan.html\nPicture: document, Americas-AI-Action-Plan.pdf, JULY 2025, Public domain, https://commons.wikimedia.org/wiki/File:Americas-AI-Action-Plan.pdf", "size": [9.8, 2.67], "position": { "x": 226.018, "y": -36.18, "z": 0 }, "features": { "resizeTextToFit": true }, "images": ["87514768-65dc-4972-8d25-d29269144fb5"] }, { "ID": "ai-boom-rg-ai-act-gpai", "name": "AI Act GPAI duties", "type": "Regulation", "properties": { "Jurisdiction": "European Union", "Regulation kind": "Law", "Status": "In force", "Date": "2025-08-02", "Impact score": 45, "URL": "https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai" }, "description": "Full name: EU AI Act: general-purpose AI model duties apply. On 2 August 2025 the EU AI Act's governance rules and obligations for providers of general-purpose AI models began to apply, according to the European Commission. The Commission had published the voluntary General-Purpose AI Code of Practice on 10 July 2025 to help providers comply. The rules cover transparency, copyright policy and, for models with systemic risk, evaluation and incident reporting, and are enforced by the AI Office. Le Monde reported on 2 August that 38 creators' organisations condemned the arrangements as a 'betrayal' of Europe's stated goals on copyright. It is the stage of the Act most directly aimed at the makers of ChatGPT-class models.\n\nhttps://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai\nhttps://www.lemonde.fr/en/economy/article/2025/08/02/ai-act-38-global-creators-organizations-condemn-betrayal-of-europe-s-stated-goals_6744006_19.html", "size": [9.8, 2.41], "position": { "x": 229.176, "y": -45.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-rg-ca-sb53", "name": "California SB 53", "type": "Regulation", "properties": { "Jurisdiction": "US state", "Regulation kind": "Law", "Status": "In force", "Date": "2025-09-29", "Impact score": 40, "URL": "https://fpf.org/blog/californias-sb-53-the-first-frontier-ai-law-explained/" }, "description": "Full name: California SB 53 Transparency in Frontier AI Act. California Governor Gavin Newsom signed Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act, on 29 September 2025. The Future of Privacy Forum, in an explainer of 3 October 2025, describes it as the first frontier-AI law: it applies to developers of foundation models trained with more than 10^26 operations, requires large developers (above $500 million revenue) to publish an annual framework on catastrophic risks and all frontier developers to publish transparency reports, requires reports of critical safety incidents to the state Office of Emergency Services within 15 days (24 hours if danger is imminent), protects whistleblowers, and lets the Attorney General seek civil penalties of up to $1 million per violation. It was the most significant US state AI safety law after federal policy turned deregulatory.\n\nhttps://fpf.org/blog/californias-sb-53-the-first-frontier-ai-law-explained/\nhttps://www.brookings.edu/articles/what-is-californias-ai-safety-law/", "size": [9.8, 2.3], "position": { "x": 245.724, "y": -52.11, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-rg-h200-china-2026", "name": "H200 export licences", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Export control", "Status": "In force", "Date": "2026-01-13", "Impact score": 50, "URL": "https://www.cov.com/en/news-and-insights/insights/2026/01/us-commerce-department-revises-license-review-policy-for-exports-of-certain-advanced-computing-commodities-to-china-and-macau" }, "description": "Full name: US allows case-by-case licences for Nvidia H200 sales to China. On 8 December 2025 President Trump announced that Nvidia's H200 and AMD's MI325X could be sold to China in exchange for a 25% payment to the US government. On 13 January 2026 the Bureau of Industry and Security issued a final rule, effective 15 January, moving licence review for exports of such chips from the US to China and Macau from a presumption of denial to case by case. Conditions include a cap at 50% of the volume shipped to US customers, third-party testing in the US, and know-your-customer certifications; chips with total processing performance below 21,000 and DRAM bandwidth below 6,500 GB/s are covered. On 14 January a separate proclamation imposed a 25% tariff on such chips. In May 2026 Senator Chris Coons asked Commerce for clearer answers on licensed shipments, and in July 2026 Under Secretary Jeffrey Kessler told Congress that shipments had been extremely rare.\n\nhttps://www.cov.com/en/news-and-insights/insights/2026/01/us-commerce-department-revises-license-review-policy-for-exports-of-certain-advanced-computing-commodities-to-china-and-macau\nhttps://pillsburylaw.com/en/news-and-insights/trump-advanced-ai-semiconductors-actions.html\nhttps://en.fnnews.com/news/202607150755449382\nhttps://news.bloomberglaw.com/artificial-intelligence/nvidia-h200-chip-exports-to-china-queried-in-letter-to-lutnick-1", "size": [9.8, 2.5], "position": { "x": 264.69, "y": -36.18, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-rg-pentagon-anthropic-risk", "name": "Anthropic risk label", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Ban or restriction", "Status": "Partly in force", "Date": "2026-03-05", "Impact score": 65, "URL": "https://fortune.com/2026/03/06/pentagon-officially-defines-anthropic-as-supply-chain-risk/" }, "description": "Full name: Pentagon designates Anthropic a supply chain risk. After Anthropic refused to drop limits on the use of Claude for mass domestic surveillance and fully autonomous weapons, Defense Secretary Pete Hegseth set a 27 February 2026 deadline and President Trump ordered federal agencies to stop using Anthropic's technology. On 5 March 2026 the Pentagon formally notified Anthropic that it and its products were a supply chain risk, effective immediately, a designation normally aimed at foreign adversaries such as Huawei; AP reported that the military was given six months to phase Claude out and that a contract worth up to $200 million was at stake. Lockheed Martin said it would follow the government's direction, Microsoft said it could keep working with Anthropic on non-defense projects, Senator Kirsten Gillibrand called it 'a dangerous misuse of a tool meant to address adversary-controlled technology', and chief executive Dario Amodei said 'we see no choice but to challenge it in court'. OpenAI announced a deal to replace Claude in classified military settings. Wikipedia's Anthropic and Claude articles report that Judge Rita Lin issued a preliminary injunction on 26 March, writing that the action 'appears to be classic First Amendment retaliation', that she ruled the blacklisting illegal in August 2026, and that BBC News reported on 5 October that the Defense Department had stopped using Anthropic products.\n\nhttps://fortune.com/2026/03/06/pentagon-officially-defines-anthropic-as-supply-chain-risk/\nhttps://en.wikipedia.org/wiki/Anthropic", "size": [9.8, 2.75], "position": { "x": 272.83, "y": -45.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-rg-commerce-anthropic-access", "name": "Mythos access curbs", "type": "Regulation", "properties": { "Jurisdiction": "United States", "Regulation kind": "Export control", "Status": "Revoked or rescinded", "Date": "2026-06-12", "Impact score": 65, "URL": "https://www.aljazeera.com/economy/2026/7/1/us-lifts-restrictions-on-powerful-ai-models-fable-mythos-anthropic-says" }, "description": "Full name: US Commerce directive restricting access to Anthropic's Mythos-class models. On 12 June 2026 Anthropic disabled access to Claude Fable 5 and Mythos 5 to comply with a US Department of Commerce directive that required it to block all foreign nationals, including its own foreign-national employees, from the models, according to Al Jazeera and Wikipedia's Claude article; BBC News headlined it as Anthropic suspending new AI tools over US government security concerns. The government gave no specific reason beyond unspecified national-security concerns, and Anthropic believed officials were worried about vulnerabilities in Fable 5. On 26 June Commerce allowed Mythos 5 for more than 100 US organisations that run critical infrastructure, and on 30 June Commerce Secretary Howard Lutnick wrote that Anthropic no longer needed an export licence, in return for Anthropic agreeing to proactively detect and address security risks, work with the government on standards for future models, and report malicious activity. Access resumed on 1 July 2026.\n\nhttps://www.aljazeera.com/economy/2026/7/1/us-lifts-restrictions-on-powerful-ai-models-fable-mythos-anthropic-says\nhttps://www.bbc.com/news/articles/c932g3v3e13o", "size": [9.8, 2.75], "position": { "x": 292.969, "y": -52.11, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-rg-eu-digital-omnibus", "name": "EU AI omnibus", "type": "Regulation", "properties": { "Jurisdiction": "European Union", "Regulation kind": "Law", "Status": "In force", "Date": "2026-07-27", "Impact score": 55, "URL": "https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai" }, "description": "Full name: EU Digital Omnibus on AI. The Digital Omnibus on AI amends the EU Artificial Intelligence Act. The European Commission proposed it on 19 November 2025 to defer the high-risk obligations; the Council and Parliament reached a provisional agreement in May 2026 (7 May according to the Commission's page, 6 May according to law firm Gibson Dunn), the Parliament confirmed it on 16 June by 423 votes to 57 with 174 abstentions (law-firm reports), and the Commission's AI Act page gives its entry into force as 27 July 2026. It moves the high-risk rules for stand-alone Annex III uses such as employment and biometrics from 2 August 2026 to 2 December 2027, and rules for AI embedded in regulated products from 2 August 2027 to 2 August 2028; gives existing systems until 2 December 2026 for the watermarking duty; adds a ninth prohibition on AI systems that generate non-consensual intimate imagery or child sexual abuse material; and strengthens the AI Office.\n\nhttps://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai\nhttps://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines/", "size": [9.8, 2.59], "position": { "x": 301.932, "y": -36.18, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-mm-alphabet-bard-demo", "name": "Alphabet falls on Bard error", "type": "Market move", "properties": { "Move kind": "One-day move", "Direction": "Fall", "Date": "2023-02-08", "Move USD bn": -100, "Move percent": -8 }, "description": "Shares of Alphabet (Nasdaq: GOOGL) fell about 8 percent on 8 February 2023, after a promotional demonstration of the Bard chatbot gave a wrong answer about the James Webb Space Telescope. Reuters reported the fall under the headline 'Google shares dive 8% after AI chatbot Bard flubs answer in ad', and Wikipedia's Bard article equates the drop to about $100 billion of market value. The error appeared a day after Microsoft showed ChatGPT-powered Bing, so investors read it as evidence that Google was behind in the race ChatGPT had started. The move is a one-day, close-to-close loss; the percentage and dollar figures are as reported by Reuters and Wikipedia, rounded.\n\nhttps://www.reuters.com/technology/google-ai-chatbot-bard-offers-inaccurate-information-company-ad-2023-02-08/\nhttps://en.wikipedia.org/wiki/Bard_(chatbot)", "size": 3.75, "position": { "x": 63.474, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-mm-chegg", "name": "Chegg halves on ChatGPT", "type": "Market move", "properties": { "Move kind": "One-day move", "Direction": "Fall", "Date": "2023-05-02", "Move USD bn": -1, "Move percent": -48 }, "description": "Shares of Chegg, the US homework-help company listed on the New York Stock Exchange, fell about 48 percent on 2 May 2023 to close at $9.08, according to the Associated Press. The evening before, chief executive Dan Rosensweig had told analysts that since March there had been 'a significant spike in student interest in ChatGPT' and that it was affecting new customer growth; revenue guidance of $175 million to $178 million for the quarter was well below the analyst consensus of $193.6 million. The Register put the drop at about 49 percent and almost $1 billion of market value. Rosensweig later called the sell-off 'extraordinarily overblown' on CNBC. It was the first widely reported case of a listed company losing value because of a chatbot.\n\nhttps://www.news4jax.com/business/2023/05/02/cheggs-stock-plunges-on-fears-of-competition-from-chatgpt/\nhttps://www.theregister.com/2023/05/02/chatgpt_chegg_ai/", "size": 1.72, "position": { "x": 48.064, "y": -61.79, "z": 0 } }, { "ID": "ai-scale-mm-nvidia-guidance-2023", "name": "Nvidia guidance jump", "type": "Market move", "properties": { "Move kind": "One-day move", "Direction": "Rise", "Date": "2023-05-25", "Move USD bn": 163, "Move percent": 24 }, "description": "On 25 May 2023 Nvidia shares rose about 24% in one session, adding roughly $163 billion in market value by WION's account, after the company reported first-quarter revenue of $7.19 billion on 24 May and forecast about $11 billion for the second quarter, against analysts' consensus of $7.2 billion. Jensen Huang said the computer industry was going through two simultaneous transitions, accelerated computing and generative AI. The Motley Fool wrote that the report revealed the surge in demand for Nvidia's chips set off by ChatGPT's launch. Intraday the stock peaked nearly 26% higher, at a record market value of $955 billion; outlets give other figures for the gain, from $163 billion to $199 billion at the peak. The ticker is NVDA on Nasdaq.\n\nhttps://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2024\nhttps://www.wionews.com/business-economy/us-chipmaker-nvidia-on-the-brink-of-crossing-1-trillion-market-value-596624\nhttps://www.fool.com/investing/2024/01/10/why-nvidia-stock-skyrocketed-239-in-2023/", "size": 4.37, "position": { "x": 54.754, "y": -61.79, "z": 0 } }, { "ID": "ai-scale-mm-nvidia-1t-2023", "name": "Nvidia hits $1 trillion", "type": "Market move", "properties": { "Move kind": "Market value milestone", "Direction": "Rise", "Date": "2023-05-30", "Move USD bn": 236, "Move percent": 31 }, "description": "On 30 May 2023 Nvidia briefly touched a market value of $1 trillion, up as much as 7.7% in the session and valued at about $1.01 trillion, according to Bloomberg, before slipping back; Yahoo Finance reported that it closed at $401.11, a market capitalisation of $992 billion. Reuters described it as the first chipmaker to join the trillion-dollar club, with TSMC next at about $535 billion. Analysts attributed the climb to AI demand after the 24 May forecast of about $11 billion in second-quarter revenue against Wall Street's $7.2 billion. From the close of 24 May, about $756 billion by WION's figures, to the close of 30 May, $992 billion by Yahoo Finance's, the gain was about $236 billion.\n\nhttps://finance.yahoo.com/news/nvidia-touches-1-trillion-market-cap-as-chipmaker-rides-ai-wave-133530381.html\nhttps://www.wionews.com/business-economy/us-chipmaker-nvidia-on-the-brink-of-crossing-1-trillion-market-value-596624\nhttps://www.businesstoday.in/amp/latest/corporate/story/nvidia-becomes-first-chipmaker-valued-at-over-1-trillion-383501-2023-05-30", "size": 4.95, "position": { "x": 20.954, "y": -61.79, "z": 0 } }, { "ID": "ai-scale-mm-nvidia-record-gain-2024", "name": "Nvidia record $277bn gain", "type": "Market move", "properties": { "Move kind": "One-day move", "Direction": "Rise", "Date": "2024-02-22", "Move USD bn": 277, "Move percent": 16.4 }, "description": "On 22 February 2024 Nvidia shares rose 16.4% to a record close of $785.38, adding $277 billion in market value, the largest one-day gain on Wall Street at the time and above Meta's $196 billion record from three weeks earlier. The value stood at about $1.96 trillion, close to $2 trillion, which Wikipedia says the company closed above on 1 March 2024. The jump followed its January-quarter report on 21 February, which beat expectations on strong demand for AI computing chips; Reuters noted that it forecast 233% growth in current-quarter revenue against 208% expected by analysts. Nvidia broke its own record again in August 2024.\n\nhttps://www.business-reporter.co.uk/news/nvidia-adds-record-277-billion-in-stock-market-value-9994\nhttps://www.ig.com/en/news-and-trade-ideas/nvidia-takes-the-record-for-the-biggest-stock-market-gain-ever-240223\nhttps://en.wikipedia.org/wiki/Nvidia", "size": 5.24, "position": { "x": 31.744, "y": -61.79, "z": 0 } }, { "ID": "ai-scale-mm-nvidia-most-valuable-2024", "name": "Nvidia most valuable", "type": "Market move", "properties": { "Move kind": "Market value milestone", "Direction": "Rise", "Date": "2024-06-18", "Move USD bn": 113, "Move percent": 3.5 }, "description": "On 18 June 2024 Nvidia shares rose 3.5% to $135.58, lifting its market value to $3.335 trillion and past Microsoft, down 0.45%, to make it the most valuable listed company, Al Jazeera reported. The report attributed the rally to voracious demand for Nvidia's chips from tech giants such as Microsoft, Meta and Google, and said the stock was up nearly 182% in 2024 and that Nvidia controlled about 80% of the market for AI chips in data centres. A 3.5% rise on the $3.335 trillion Al Jazeera reported means a gain of about $113 billion that day. Reuters-based figures differ slightly, at $3.326 trillion.\n\nhttps://www.aljazeera.com/economy/2024/6/19/nvidia-becomes-worlds-most-valuable-company-dethroning-microsoft\nhttps://business.dptribune.com/dptribune/news/read/44159210/cnbc_daily_open", "size": 3.89, "position": { "x": 156.363, "y": -61.79, "z": 0 } }, { "ID": "ai-a-nvidia-deepseek-selloff", "name": "Nvidia DeepSeek selloff", "type": "Market move", "properties": { "Move kind": "One-day move", "Direction": "Fall", "Date": "2025-01-27", "Move USD bn": -593, "Move percent": -17 }, "description": "On 27 January 2025 Nvidia shares fell just under 17% and the company lost $593 billion of market value, a record one-day loss for a Wall Street company according to Reuters, which cited LSEG data and noted it was more than double the previous record, also set by Nvidia, in September 2024. Investors worried that a low-cost Chinese AI model from DeepSeek, whose free assistant had overtaken ChatGPT in Apple's App Store downloads, would threaten the dominance of AI leaders. The Nasdaq fell 3.1%, the Philadelphia semiconductor index 9.2%, Broadcom 17.4%, Marvell 19.1%, SoftBank 8.3% in Japan and ASML 7% in Europe. The loss is given as $593 billion by Reuters and $589 billion in the same report's photo caption; Wikipedia says about $600 billion and an 18% price fall.\n\nhttps://www.koreajoongangdaily.com/business/investors-dump-tech-stocks-after-record-593-billion-nvidia-loss/12473281\nhttps://en.wikipedia.org/wiki/DeepSeek\nhttps://en.wikipedia.org/wiki/Nvidia", "size": 6.97, "position": { "x": 180.48, "y": -61.79, "z": 0 } }, { "ID": "ai-scale-mm-nvidia-4t-2025", "name": "Nvidia hits $4 trillion", "type": "Market move", "properties": { "Move kind": "Market value milestone", "Direction": "Rise", "Date": "2025-07-09", "Move USD bn": 109, "Move percent": 2.8 }, "description": "On 9 July 2025 Nvidia became the first listed company to reach a market value of $4 trillion. Bloomberg reported that shares rose 2.8% to $164.42, passing Apple and Microsoft, and Business Today attributed the climb to demand for the chips that power the data centres behind AI models at Microsoft, Amazon and Google. Reuters-based reports said the value held only briefly, with the close at about $3.97 trillion, and Wikipedia gives 10 July as the first close above $4 trillion, so sources differ on whether the milestone was an intraday or closing one. Bloomberg's 2.8% rise to the $4 trillion level means a gain of about $109 billion that day.\n\nhttps://www.businesstoday.in/tech-today/news/story/nvidia-ai-chip-demand-drives-4-trillion-market-value-483958-2025-07-09\nhttps://www.bloomberg.com/news/articles/2025-07-09/nvidia-hits-4-trillion-value-as-rally-notches-another-milestone\nhttps://en.wikipedia.org/wiki/Nvidia", "size": 3.84, "position": { "x": 138.573, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-mm-oracle-openai", "name": "Oracle surges on OpenAI", "type": "Market move", "properties": { "Move kind": "One-day move", "Direction": "Rise", "Date": "2025-09-10", "Move USD bn": 234, "Move percent": 36.7 }, "description": "Oracle (NYSE: ORCL) shares rose about 36.7 percent late in the session on 10 September 2025, after hitting a record intraday gain of about 43 percent, adding roughly $234 billion of market value and taking its valuation to about $913 billion, Reuters reported. It was Oracle's biggest one-day percentage gain since 1992. The jump followed Oracle's disclosure of four multi-billion-dollar contracts and a Wall Street Journal report that OpenAI had signed a $300 billion, five-year computing deal with it. Forbes estimated that Larry Ellison's fortune rose by about $100 billion to roughly $392.6 billion, briefly putting him ahead of Elon Musk. The figures are Reuters' late-session numbers and may differ slightly from the official close.\n\nhttps://www.tbsnews.net/node/1233721\nhttps://about.getcoai.com/?p=83484", "size": 4.93, "position": { "x": 239.502, "y": -74.29, "z": 0 } }, { "ID": "ai-scale-mm-amd-openai-2025", "name": "AMD jumps on OpenAI deal", "type": "Market move", "properties": { "Move kind": "One-day move", "Direction": "Rise", "Date": "2025-10-06", "Move USD bn": 63.4, "Move percent": 23.7 }, "description": "On 6 October 2025 AMD shares surged after the company and OpenAI announced a multi-year agreement under which OpenAI will deploy 6 gigawatts of AMD GPUs, starting with 1 gigawatt of Instinct MI450 chips in the second half of 2026. AMD issued OpenAI a warrant for up to 160 million shares, about 10% of the company, at a penny each, vesting as deployment and share-price milestones are met, the last at $600. Forbes reported gains of more than 30% to about $215 at the open and ChannelNews 27%; a Dow Jones-based report summarised by Advisor Perspectives put the close 23.7% higher at $203.71, a one-day gain of $63.36 billion to a market value of $330.6 billion. AMD said it expected tens of billions of dollars in revenue from the deal.\n\nhttps://www.forbes.com.au/?p=159936\nhttps://www.channelnews.com.au/amd-challenges-nvidia-dominance-with-landmark-openai-chip-agreement/\nhttps://www.advisorperspectives.com/articles/2025/10/06/amd-soars-openai-deal-worth-billions\nhttps://www.fool.com/investing/2025/10/06/amd-stock-skyrockets-on-massive-deal-with-openai-c", "size": 3.29, "position": { "x": 261.106, "y": -61.79, "z": 0 } }, { "ID": "ai-scale-mm-nvidia-5t-2025", "name": "Nvidia hits $5 trillion", "type": "Market move", "properties": { "Move kind": "Market value milestone", "Direction": "Rise", "Date": "2025-10-29", "Move USD bn": 170, "Move percent": 3.4 }, "description": "On 29 October 2025 Nvidia became the first company valued at $5 trillion. Forbes reported that its market value rose from about $4.89 trillion to $5.06 trillion as shares gained 3.4% to around $207.85 at the open, ahead of Microsoft at $4 trillion and Apple at $3.9 trillion; Entrepreneur put the shares above $210, up over 4%, and said the $5 trillion threshold needed a share price of $205.76. Both attributed the rise to Jensen Huang's announcement a day earlier that Nvidia expects about $500 billion in AI chip orders and will build seven supercomputers for the US government, and to a $1 billion stake in Nokia. The gain of about $170 billion is the intraday value change in Forbes's figures.\n\nhttps://www.forbes.com.au/news/investing/nvidia-becomes-first-company-worth-5-trillion/\nhttps://www.entrepreneur.com/business-news/nvidia-becomes-the-first-to-pass-5-trillion-in-market-value", "size": 4.43, "position": { "x": 147.443, "y": -61.79, "z": 0 } }, { "ID": "ai-a-fifth-generation-project", "name": "Fifth Generation project", "type": "Funding round", "properties": { "Deal kind": "Government programme", "Date": "1982-01-01", "Amount USD bn": 0.32 }, "description": "Japan's Ministry of International Trade and Industry (MITI) launched the Fifth Generation Computer Systems project in 1982 and set up the Institute for New Generation Computer Technology (ICOT) to run it. The aim was a ten-year plan (three years of research, four of subsystems, three for a prototype) to build parallel computers for AI based on concurrent logic programming, targeting 100 million to a billion logical inferences per second, against about 100,000 for typical workstations. The project's own final evaluation, cited on Wikipedia, puts the total spend at a little under 57 billion yen, about 320 million dollars; Wikipedia's AI winter article gives a larger figure, 850 million dollars set aside by MITI in 1981. The project produced five parallel inference machines and contributions to concurrent logic programming, but was a commercial failure, ending in 1992 by one account and in 1994 by another, and it spurred the US Strategic Computing Initiative, Britain's Alvey programme and Europe's ESPRIT.\n\nhttps://en.wikipedia.org/wiki/Fifth_generation_computer\nhttps://en.wikipedia.org/wiki/AI_winter\nhttps://www.ueda.info.waseda.ac.jp/AITEC_ICOT_ARCHIVES/ICOT/Museum/FinalReport/final.html", "size": 1.71, "position": { "x": -229.195, "y": -61.79, "z": 0 } }, { "ID": "ai-a-strategic-computing", "name": "Strategic Computing", "type": "Funding round", "properties": { "Deal kind": "Government programme", "Date": "1983-01-01", "Amount USD bn": 1 }, "description": "The Strategic Computing Initiative was a DARPA programme begun in 1983 that aimed at machine intelligence able to 'see, hear, speak, and think like a human'. Directed by the Information Processing Technology Office, it spent a total of about one billion dollars; by 1985 it had spent 100 million dollars on 92 projects at 60 institutions. It was launched in the shadow of Japan's Fifth Generation project and built on goals such as autonomous land navigation. The high-level goal was not reached, but the Autonomous Land Vehicle and Navlab work laid groundwork for later driverless-vehicle programmes and the logistics planning tool DART was a success in Desert Storm. In 1987 and 1988 DARPA cut AI funding sharply and cancelled new AI spending under the programme; it ended in 1993.\n\nhttps://en.wikipedia.org/wiki/Strategic_Computing_Initiative\nhttps://en.wikipedia.org/wiki/AI_winter", "size": 1.72, "position": { "x": -223.566, "y": -61.79, "z": 0 } }, { "ID": "ai-hist-fr-google-dnnresearch", "name": "Google acquires DNNresearch", "type": "Funding round", "properties": { "Deal kind": "Acquisition", "Date": "2013-03-12", "Amount USD bn": 0.044 }, "description": "On 12 March 2013 Google acquired DNNresearch, the Toronto start-up founded the year before by Geoffrey Hinton, Alex Krizhevsky and Ilya Sutskever, months after AlexNet's win in the ImageNet challenge. Google confirmed the acquisition but disclosed no terms; InformationWeek and the IDG News Service both reported that the price was not given. Hinton later said the company had been put up for auction among several bidders, and in Cade Metz's book Genius Makers, as relayed in an excerpt, he stopped the bidding when the price reached 44 million dollars, so the amount below is a reported, not confirmed, figure. The deal brought the three researchers into Google, where Sutskever joined Google Brain, and was an early step in the corporate race for deep learning talent.\n\nhttps://www.informationweek.com/data-management/google-buys-machine-learning-startup\nhttps://www.infoworld.com/article/2267586/google-acquires-startup-focused-on-neural-networks.html\nhttps://www.goodreads.com/notes/57473123-genius-makers/83395142-ad", "size": 1.55, "position": { "x": -145.465, "y": -61.79, "z": 0 } }, { "ID": "ai-a-deepmind-acquisition", "name": "Google acquires DeepMind", "type": "Funding round", "properties": { "Deal kind": "Acquisition", "Date": "2014-01-26", "Amount USD bn": 0.5 }, "description": "On 26 January 2014 Google agreed to buy DeepMind, the London AI company founded in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman. Re/code first reported a price of 400 million dollars, which Google declined to confirm; The Information, as relayed by TechCrunch, put it at more than 500 million dollars and said Facebook had also tried to buy the company late the previous year, and Wikipedia summarises reported prices of 400 to 650 million dollars. DeepMind confirmed the deal without terms. According to The Information's sources Google agreed to set up an ethics board to oversee DeepMind's technology, whose members were never disclosed. The purchase gave Google the team behind the Atari-playing Deep Q-network and later AlphaGo, and the company merged with Google Brain in April 2023.\n\nhttps://techcrunch.com/2014/01/26/google-deepmind/\nhttps://www.nbcnews.com/id/wbna54190647\nhttps://en.wikipedia.org/wiki/Google_DeepMind", "size": 1.66, "position": { "x": -128.186, "y": -61.79, "z": 0 } }, { "ID": "ai-hist-fr-openai-founding", "name": "OpenAI founding pledge", "type": "Funding round", "properties": { "Deal kind": "Strategic investment", "Date": "2015-12-11", "Amount USD bn": 1 }, "description": "OpenAI was announced on 11 December 2015 as a non-profit AI research company with Sam Altman and Elon Musk as co-chairs and Ilya Sutskever as research director. The backers named were Altman, Greg Brockman, Musk, Jessica Livingston, Peter Thiel, Amazon Web Services, Infosys and YC Research, and TechCrunch reported that they had committed about 1 billion dollars. Later accounts show the pledge was larger than the cash: Wikipedia, citing tax filings, says only 133 million dollars had been received by 2021. The stated mission was to advance digital intelligence in the way most likely to benefit humanity as a whole, free of any need to generate financial return. The money recruited researchers away from Google and became the base of what turned into the largest AI company of the 2020s.\n\nhttps://techcrunch.com/2015/12/11/non-profit-openai-launches-with-backing-from-elon-musk-and-sam-altman\nhttps://en.wikipedia.org/wiki/OpenAI", "size": 1.72, "position": { "x": -101.052, "y": -61.79, "z": 0 } }, { "ID": "ai-hist-fr-south-korea-ai-fund", "name": "Korea AI fund 2016", "type": "Funding round", "properties": { "Deal kind": "Government programme", "Date": "2016-03-17", "Amount USD bn": 0.863 }, "description": "On 17 March 2016, two days after AlphaGo's 4 to 1 victory over Lee Sedol in Seoul, the South Korean government announced that it would invest about 1 trillion won, roughly 863 million dollars, in artificial intelligence research over five years. The plan included a public-private research centre with companies such as Samsung, LG Electronics, Hyundai Motor and Naver. It was unclear whether the money was new or previously allocated, and academics complained that the funds favoured industry over universities. President Park Geun-hye called AI 'the fourth industrial revolution'. Nature's news service headlined its report 'South Korea trumpets $860 million AI fund after AlphaGo shock'.\n\nhttps://en.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol\nhttps://behindtheblack.com/?p=38271\nhttps://www.nature.com/news/south-korea-trumpets-860-million-ai-fund-after-alphago-shock-1.19595", "size": 1.71, "position": { "x": -96.348, "y": -61.79, "z": 0 } }, { "ID": "ai-a-microsoft-openai-2019", "name": "Microsoft $1bn in OpenAI", "type": "Funding round", "properties": { "Deal kind": "Strategic investment", "Date": "2019-07-22", "Amount USD bn": 1 }, "description": "On 22 July 2019 Microsoft and OpenAI announced a multiyear partnership in which Microsoft would invest $1 billion in OpenAI. OpenAI said it would port its services to Microsoft Azure, and Microsoft would be its preferred partner for commercialising new technologies; the two said they would work on Azure supercomputing for AI. Sam Altman said the creation of artificial general intelligence would be the most important technological development in human history, and Satya Nadella called AI one of the most transformative technologies of our time. The money arrived months after OpenAI created a capped-profit arm, and it paid for the compute that trained GPT-3 on a Microsoft-provided cluster.\n\nhttps://news.microsoft.com/2019/07/22/openai-forms-exclusive-computing-partnership-with-microsoft-to-build-new-azure-ai-supercomputing-technologies/\nhttps://en.wikipedia.org/wiki/OpenAI", "size": 1.72, "position": { "x": -40.335, "y": -61.79, "z": 0 } }, { "ID": "ai-scale-fr-nvidia-arm", "name": "Nvidia bid for Arm", "type": "Funding round", "properties": { "Deal kind": "Acquisition", "Date": "2020-09-13", "Amount USD bn": 40 }, "description": "On 13 September 2020 Nvidia agreed to acquire the chip designer Arm from SoftBank for $40 billion in stock and cash, the largest semiconductor acquisition to that date; SoftBank would have taken a stake of slightly less than 10% in Nvidia. The UK raised national security concerns, Google, Microsoft and Qualcomm raised competition concerns because their chips rely on Arm designs, and the European Commission, the UK Competition and Markets Authority and the US Federal Trade Commission all questioned the deal. Nvidia and SoftBank cancelled it in February 2022 under regulatory pressure, and SoftBank took Arm public on Nasdaq in September 2023.\n\nhttps://nvidianews.nvidia.com/news/nvidia-to-acquire-arm-for-40-billion-creating-worlds-premier-computing-company-for-the-age-of-ai\nhttps://www.theguardian.com/business/2022/feb/08/nvidia-takeover-arm-collapses-softbank\nhttps://en.wikipedia.org/wiki/Arm_Holdings", "size": 2.92, "position": { "x": -13.417, "y": -61.79, "z": 0 } }, { "ID": "ai-scale-fr-stability-101m", "name": "Stability AI $101m", "type": "Funding round", "properties": { "Deal kind": "Equity round", "Date": "2022-10-17", "Amount USD bn": 0.101 }, "description": "On 17 October 2022 Stability AI, two months after releasing Stable Diffusion, announced a $101 million funding round led by Coatue and Lightspeed Venture Partners, with O'Shaughnessy Ventures also taking part, at a reported valuation of $1 billion. TechCrunch reported the round on the day, and Forbes had written on 7 September that the company was in talks to raise at a valuation of up to $1 billion. The money funded more compute and staff for a company whose model had been downloaded freely; within a year Lightspeed and Coatue were questioning its management, which led to its CEO's resignation in March 2024.\n\nhttps://techcrunch.com/2022/10/17/stability-ai-the-startup-behind-stable-diffusion-raises-101m/\nhttps://en.wikipedia.org/wiki/Stability_AI\nhttps://www.forbes.com/sites/kenrickcai/2022/09/07/stability-ai-funding-round-1-billion-valuation-stable-diffusion-text-to-image/", "size": 1.57, "position": { "x": 39.154, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-microsoft-openai-2023", "name": "Microsoft $10bn in OpenAI", "type": "Funding round", "properties": { "Deal kind": "Strategic investment", "Date": "2023-01-23", "Amount USD bn": 10 }, "description": "On 23 January 2023 Microsoft announced a multiyear, multibillion-dollar extension of its partnership with OpenAI, following its 2019 and 2021 investments. The announcement gave no dollar figure; the widely repeated 'about $10 billion' came from earlier press reports in January 2023, and Wikipedia's OpenAI article gives $10 billion over several years, mostly as Azure compute, at a roughly $29 billion valuation. Microsoft said it would be the exclusive cloud provider for OpenAI, deploy OpenAI models across its products and invest in specialised supercomputing systems. The deal came weeks after ChatGPT's launch, put OpenAI's models into Microsoft's Bing and Office products, and became a pillar of the Microsoft-OpenAI relationship that was restructured in October 2025. The amount used for size here is the reported figure, not an announced one.\n\nhttps://onmsft.com/news/microsoft-extends-partnership-with-openai/\nhttps://en.wikipedia.org/wiki/OpenAI", "size": 2.21, "position": { "x": 43.534, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-amazon-anthropic-2023", "name": "Amazon $4bn in Anthropic", "type": "Funding round", "properties": { "Deal kind": "Strategic investment", "Date": "2023-09-01", "Amount USD bn": 4 }, "description": "In September 2023 Amazon announced a partnership with Anthropic under which it would invest up to $4 billion, beginning with $1.25 billion; the remaining $2.75 billion followed in March 2024, according to Wikipedia's Anthropic article and Associated Press coverage of the later deal. Anthropic agreed to use Amazon Web Services as a main cloud provider and Amazon's custom chips. It was the first of several Amazon investments, which reached $8 billion with a further $4 billion in November 2024.\n\nhttps://en.wikipedia.org/wiki/Anthropic\nhttps://abcnews.go.com/US/wireStory/amazon-invest-additional-4-billion-ai-startup-anthropic-116134356", "size": 1.95, "position": { "x": 100.859, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-openai-6-6bn", "name": "OpenAI $6.6bn round", "type": "Funding round", "properties": { "Deal kind": "Equity round", "Date": "2024-10-02", "Amount USD bn": 6.6 }, "description": "On 2 October 2024 OpenAI closed a round that raised $6.6 billion in a funding round that valued it at $157 billion, with Microsoft, Nvidia and SoftBank taking part, as Reuters reported. It followed the January 2023 Microsoft deal and the November 2023 governance crisis and was the largest venture round to that date. Further rounds followed in 2025 and 2026 at $300 billion, $500 billion (secondary sale) and, in February 2026, $730 billion before the money.\n\nhttps://en.wikipedia.org/wiki/OpenAI\nhttps://www.reuters.com/technology/artificial-intelligence/openai-closes-66-billion-funding-haul-valuation-157-billion-with-investment-2024-10-02/", "size": 2.08, "position": { "x": 162.933, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-amazon-anthropic-2024", "name": "Amazon adds $4bn", "type": "Funding round", "properties": { "Deal kind": "Strategic investment", "Date": "2024-11-22", "Amount USD bn": 4 }, "description": "On 22 November 2024 Amazon announced an additional $4 billion investment in Anthropic, bringing its total to $8 billion, with Anthropic naming Amazon Web Services its primary cloud and training partner. The Associated Press, NBC News and Bloomberg reported that Amazon would remain a minority investor. The investment followed the $4 billion committed in 2023 and 2024 and tied Anthropic's training to Amazon's Trainium chips, setting up a rivalry with the Microsoft-OpenAI pairing.\n\nhttps://abcnews.go.com/US/wireStory/amazon-invest-additional-4-billion-ai-startup-anthropic-116134356\nhttps://www.nbcnews.com/business/business-news/amazon-invest-another-4-billion-anthropic-openais-biggest-rival-rcna181360", "size": 1.95, "position": { "x": 170.96, "y": -61.79, "z": 0 } }, { "ID": "ai-a-stargate", "name": "Stargate", "type": "Funding round", "properties": { "Deal kind": "Infrastructure commitment", "Date": "2025-01-21", "Amount USD bn": 500 }, "description": "Stargate is a US AI-infrastructure joint venture announced on 21 January 2025 at the White House by President Donald Trump with Sam Altman of OpenAI, Larry Ellison of Oracle and Masayoshi Son of SoftBank. Reuters reported the headline as a plan to invest $500 billion in AI infrastructure; Wikipedia gives an initial $100 billion with a target of $500 billion by 2029, with SoftBank and OpenAI each reported to contribute $19 billion and Oracle and MGX $7 billion each. Masayoshi Son is chairman. The first data centres are in Abilene, Texas; in September 2025 OpenAI said the venture had nearly 7 gigawatts of planned capacity and over $400 billion of investment. A Bloomberg report in August 2025 said the project had not yet raised the funds for its initial budget, and by May 2026 a planned Argentine site had advanced no further than a letter of intent. The $500 billion figure is a commitment ceiling, not money paid in.\n\nhttps://www.reuters.com/technology/artificial-intelligence/openai-softbank-oracle-invest-500-bln-ai-trump-says-2025-01-21/\nhttps://en.wikipedia.org/wiki/Stargate_LLC", "size": 6.52, "position": { "x": 194.57, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-softbank-openai-40bn", "name": "OpenAI $40bn round", "type": "Funding round", "properties": { "Deal kind": "Equity round", "Date": "2025-03-01", "Amount USD bn": 40 }, "description": "In spring 2025 OpenAI raised $40 billion at a $300 billion post-money valuation in a round led by SoftBank, according to Wikipedia's OpenAI article; the brief for this record dates it to March 2025 while Wikipedia gives April, so the date carries month precision. SoftBank later approved a second instalment of $22.5 billion in October 2025 to complete a $30 billion commitment, contingent on OpenAI completing its restructuring, per Wikipedia's SoftBank article. At the time it was the largest private funding round on record; Crunchbase later ranked Anthropic's $30 billion Series G of February 2026 as the second largest ever, behind OpenAI's $40 billion round of 2025.\n\nhttps://en.wikipedia.org/wiki/OpenAI\nhttps://en.wikipedia.org/wiki/SoftBank_Group", "size": 2.92, "position": { "x": 204.61, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-meta-scale-ai", "name": "Meta $14.3bn in Scale AI", "type": "Funding round", "properties": { "Deal kind": "Strategic investment", "Date": "2025-06-12", "Amount USD bn": 14.3 }, "description": "In June 2025 Meta agreed to invest $14.3 billion for a 49 percent non-voting stake in Scale AI, the data-labelling company, and to hire its chief executive Alexandr Wang as Meta's chief AI officer. Reuters reported the talks on 10 June, Fortune and The New York Times reported the completed investment on 12-13 June. The deal opened Meta's 'superintelligence' hiring wave, in which Mark Zuckerberg personally recruited researchers with pay packages The New York Times reported at between $1 million and $100 million; Meta formally created Meta Superintelligence Labs on 30 June 2025.\n\nhttps://fortune.com/2025/06/13/meta-scale-ai-alexandr-wang-superintelligence-team\nhttps://en.wikipedia.org/wiki/Meta_Superintelligence_Labs", "size": 2.35, "position": { "x": 214.397, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-anthropic-13bn", "name": "Anthropic $13bn round", "type": "Funding round", "properties": { "Deal kind": "Equity round", "Date": "2025-09-02", "Amount USD bn": 13 }, "description": "On 2 September 2025 Anthropic announced a $13 billion Series F at a $183 billion post-money valuation, led by Iconiq with Fidelity Management and Research and Lightspeed Venture Partners as co-leads and participation from the Qatar Investment Authority, Blackstone and Coatue, according to Anthropic's announcement as summarised in press reports. The previous round had been $3.5 billion at $61.5 billion in March 2025. Anthropic said it would use the money for enterprise adoption, safety research and international expansion, and its announcement put its run-rate revenue at over $5 billion by August 2025, up from about $1 billion at the start of the year.\n\nhttps://finance.yahoo.com/news/anthropic-secures-13b-series-f-094048133.html\nhttps://en.wikipedia.org/wiki/Anthropic", "size": 2.31, "position": { "x": 235.426, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-oracle-openai-300bn", "name": "OpenAI-Oracle $300bn deal", "type": "Funding round", "properties": { "Deal kind": "Infrastructure commitment", "Date": "2025-09-10", "Amount USD bn": 300 }, "description": "On 10 September 2025 the Wall Street Journal reported that OpenAI had signed a contract to buy $300 billion of computing power from Oracle over about five years, following Oracle's disclosure the day before of four multi-billion-dollar contracts. Neither company confirmed the figure at the time; Wikipedia's OpenAI article later gives a $300 billion, five-year Oracle deal, and Wikipedia's Oracle article lists a $300 billion, 4.5-gigawatt contract signed in September 2025. The deal was part of the Stargate build-out and, with Nvidia's letter of intent twelve days later, marked the shift of AI financing toward long-term compute commitments. Its size is a headline commitment, not paid-in capital, and some reports said the contract would begin in 2027.\n\nhttps://www.tbsnews.net/node/1233721\nhttps://en.wikipedia.org/wiki/OpenAI", "size": 5.39, "position": { "x": 243.726, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-nvidia-openai-100bn", "name": "Nvidia $100bn OpenAI plan", "type": "Funding round", "properties": { "Deal kind": "Infrastructure commitment", "Date": "2025-09-22", "Amount USD bn": 100 }, "description": "On 22 September 2025 Nvidia and OpenAI announced a letter of intent under which Nvidia plans to invest up to $100 billion in OpenAI as at least 10 gigawatts of Nvidia systems are deployed, the first gigawatt on the Vera Rubin platform in the second half of 2026. Nvidia's press release said the money would arrive progressively with each gigawatt; a Reuters-sourced report said the first $10 billion would follow a definitive agreement for chip purchases. The deal was a letter of intent, not a binding contract, and critics described it as 'circular financing' because OpenAI would spend the money on Nvidia chips (Business Standard, 24 September 2025). Wikipedia's OpenAI article says the 10-gigawatt plan was later scaled back and amended in August 2026 to a leased 8-gigawatt site in Ohio with a $105 billion Nvidia backstop, and in February 2026 Nvidia committed $30 billion to OpenAI's larger round.\n\nhttps://www.bnnbloomberg.ca/business/2025/09/22/nvidia-to-invest-us100-billion-in-openai/\nhttps://www.barchart.com/story/news/34958327/openai-and-nvidia-announce-strategic-partnership-to-deploy-10-gigawatts-of-nvidia-systems", "size": 3.75, "position": { "x": 253.466, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-disney-openai", "name": "Disney $1bn in OpenAI", "type": "Funding round", "properties": { "Deal kind": "Strategic investment", "Date": "2025-12-11", "Amount USD bn": 1 }, "description": "On 11 December 2025 The Walt Disney Company announced a $1 billion investment in OpenAI and a three-year licensing agreement allowing users to generate videos with more than 200 Disney, Marvel, Pixar and Star Wars characters on Sora 2, CNBC and Disney reported. The deal tied one of the largest owners of film characters to a generative-video service at a time when studios were suing other AI companies. On 24 March 2026 OpenAI announced it would discontinue Sora, and Deadline headlined that Disney's investment was 'DOA'; a Disney spokesperson told CBS News the companies had a constructive collaboration and Disney would keep engaging with AI platforms while respecting intellectual property.\n\nhttps://www.cnbc.com/2025/12/11/disney-openai-sora-characters-video.html\nhttps://www.cbsnews.com/news/sora-ai-openai-discontinues/", "size": 1.72, "position": { "x": 130.016, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-anthropic-30bn", "name": "Anthropic $30bn round", "type": "Funding round", "properties": { "Deal kind": "Equity round", "Date": "2026-02-12", "Amount USD bn": 30 }, "description": "On 12 February 2026 Anthropic announced a $30 billion Series G at a $380 billion post-money valuation, led by GIC and Coatue, with D.E. Shaw Ventures, Dragoneer, Founders Fund, Iconiq and MGX as co-leads and including previously announced investments by Microsoft and Nvidia. Crunchbase reported it as the largest venture round of 2026 so far and the second-largest ever, behind OpenAI's $40 billion round of 2025. Anthropic said the money would fund frontier research, product development and infrastructure. Three months later Anthropic raised a further $65 billion at $965 billion.\n\nhttps://news.crunchbase.com/?p=93138\nhttps://seekingalpha.com/news/4551396-anthropic-raises-30b-at-380b-valuation-includes-money-from-microsoft-nvidia-others", "size": 2.73, "position": { "x": 268.632, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-openai-110bn", "name": "OpenAI $110bn round", "type": "Funding round", "properties": { "Deal kind": "Equity round", "Date": "2026-02-27", "Amount USD bn": 110 }, "description": "On 27 February 2026 OpenAI announced a funding round of $110 billion at a $730 billion pre-money valuation, about $840 billion after the money, with Amazon committing $50 billion and Nvidia and SoftBank $30 billion each. Bloomberg reported that Amazon would pay $15 billion at first and $35 billion more when conditions were met, reportedly an IPO or OpenAI declaring it had reached artificial general intelligence; the deal includes 2 gigawatts of computing on Amazon's Trainium chips and makes AWS the exclusive third-party cloud provider for OpenAI's enterprise agent platform Frontier. Bloomberg's source said OpenAI expected about $10 billion more from venture and sovereign funds. Wikipedia's OpenAI article says the round was extended to $120 billion in March and closed at $122 billion committed and an $852 billion post-money valuation in April, so sources differ on the final figures. The commitments include conditional and staged tranches, not cash received.\n\nhttps://finance.yahoo.com/news/chatgpt-maker-openai-receives-110b-151748215.html\nhttps://www.advisorperspectives.com/articles/2026/02/28/openai-finalizes-110-billion-730-billion-value", "size": 3.85, "position": { "x": 275.812, "y": -61.79, "z": 0 } }, { "ID": "ai-boom-fr-anthropic-65bn", "name": "Anthropic $65bn round", "type": "Funding round", "properties": { "Deal kind": "Equity round", "Date": "2026-05-28", "Amount USD bn": 65 }, "description": "On 28 May 2026 Anthropic announced a $65 billion round at a $965 billion post-money valuation, which Reuters and Bloomberg reported as surpassing OpenAI, whose round had closed in April at $852 billion. Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital led it, each reportedly putting in more than $2 billion; Amazon added $5 billion under an earlier commitment, Google several billion, and the chipmakers Micron, Samsung Electronics and SK Hynix also invested undisclosed amounts. Anthropic said it would use the funds for safety and interpretability research, compute and products. The round exceeded Anthropic's original $30 billion target and came days before its confidential IPO filing on 1 June 2026.\n\nhttps://news.bloomberglaw.com/private-equity/anthropic-raises-at-965-billion-valuation-eclipsing-openai-1\nhttps://www.outlookbusiness.com/amp/story/corporate/anthropic-surpasses-openai-with-965-bn-valuation-after-latest-fundraise", "size": 3.31, "position": { "x": 290.728, "y": -61.79, "z": 0 } }, { "ID": "ai-hist-na-chessbase-cheating-move-2015", "name": "Cheating-move analysis", "type": "News article", "properties": { "Publisher": "ChessBase", "Article kind": "Analysis", "Headline": "Deep Blue's cheating move", "Date": "2015-02-19", "URL": "https://en.chessbase.com/post/deep-blue-s-cheating-move" }, "description": "ChessBase News published this analysis by Albert Silver on 19 February 2015, eighteen years after the match, to test the move in Game 2 that led Garry Kasparov to accuse the Deep Blue team of cheating. It quotes grandmaster Yasser Seirawan, who told Wired in 2001 it was 'an incredibly refined move', and John Nunn's annotation calling it a cruel move that cut the rope which might have saved the drowning Kasparov. The author does not endorse the accusation: he shows that modern engines such as Komodo 8 choose the same move and that another winning path existed, so the move alone is not evidence of human help.\n\nhttps://en.chessbase.com/post/deep-blue-s-cheating-move", "size": [9.8, 1.5], "position": { "x": -111.517, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-na-koreatimes-lee-beatable-2016", "name": "Lee shows AlphaGo beatable", "type": "News article", "properties": { "Publisher": "The Korea Times", "Article kind": "News report", "Headline": "Lee Se-dol shows AlphaGo beatable", "Date": "2016-03-13", "URL": "https://www.koreatimes.co.kr/www/news/tech/2016/03/133_200267.html" }, "description": "The Korea Times reported on 13 March 2016, in a story by Yoon Sung-won, that Lee Se-dol had beaten AlphaGo in Game 4 of the five-game match, winning by resignation after 180 moves and ending a three-game losing streak, although AlphaGo had already won the match. Commentators said AlphaGo made strange moves in the middle of the game that let Lee take control; Lee asked to play black in the final game and DeepMind chief executive Demis Hassabis agreed. It quotes Lee calling the win priceless, Hassabis saying that Lee was too strong for AlphaGo that day, and the nine-dan professional Song Tae-gon predicting that the odd play could be the start of AlphaGo's self-destruction.\n\nhttps://www.koreatimes.co.kr/www/news/tech/2016/03/133_200267.html", "size": [9.8, 1.5], "position": { "x": -96.348, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-na-joongang-alphago-shock-2016", "name": "The AlphaGo shock", "type": "News article", "properties": { "Publisher": "Korea JoongAng Daily", "Article kind": "Opinion", "Headline": "The AlphaGo shock", "Date": "2016-03-16", "URL": "https://www.koreajoongangdaily.com/opinion/the-alphago-shock/11510655" }, "description": "The JoongAng Ilbo editorial 'The AlphaGo shock', published in print on 16 March 2016 and online in the Korea JoongAng Daily that evening, says that AlphaGo's 4 to 1 win over the Korean champion Lee Se-dol shows AI may soon outperform humans in games and take over tasks in other fields sooner than expected. It argues that Korea is falling behind because its economy relies on hardware such as memory chips and undervalues software, including through restrictions on the gaming industry, and calls for long-term planning, law changes to speed up R&D investment and public education about the coming AI era. It refers indirectly to Demis Hassabis's goal of building AI that can be applied across many areas.\n\nhttps://www.koreajoongangdaily.com/opinion/the-alphago-shock/11510655", "size": [9.8, 1.5], "position": { "x": -93.413, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-hist-na-foreignpolicy-alphago-2016", "name": "AlphaGo and civilizations", "type": "News article", "properties": { "Publisher": "Foreign Policy", "Article kind": "News report", "Headline": "AlphaGo and the Clash of Civilizations", "Date": "2016-03-18", "URL": "https://foreignpolicy.com/2016/03/18/china-go-chess-west-east-technology-artificial-intelligence-google/" }, "description": "Foreign Policy's Tea Leaf Nation channel published this report by Yifu Dong on 18 March 2016, after AlphaGo's 4 to 1 win, on how the result was debated in China. It gathers views from Go players, internet users and industry figures: some saw a contest between Eastern wisdom and Western technology and others rejected that framing, some Weibo users used the result to criticise Baidu's lag behind Google and China's restrictions on Google, and the article notes Google's choice of South Korea as the venue and the rise of Ke Jie, who might face AlphaGo next. It quotes the Beijing AI start-up chief Chen Lei, the professionals Sun Yuan, Xu Ying, Zhou Junxun and Chang Hao, the national team head Hua Xueming, Lee Sedol and, via a Twitter post, Demis Hassabis.\n\nhttps://foreignpolicy.com/2016/03/18/china-go-chess-west-east-technology-artificial-intelligence-google/", "size": [9.8, 1.5], "position": { "x": -89.192, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-verge-gpt2-debate", "name": "Ethics of sharing AI", "type": "News article", "properties": { "Publisher": "The Verge", "Article kind": "News report", "Headline": "AI researchers debate the ethics of sharing potentially harmful programs", "Date": "2019-02-21", "URL": "https://www.theverge.com/2019/2/21/18234500/ai-ethics-debate-researchers-harmful-programs-openai" }, "description": "The Verge's James Vincent reported on 21 February 2019 on the argument that followed OpenAI's decision to withhold the full GPT-2 model. The piece collected reactions from the research community on whether labs should hold back potentially harmful AI systems and whether OpenAI's warning was proportionate.\n\nhttps://www.theverge.com/2019/2/21/18234500/ai-ethics-debate-researchers-harmful-programs-openai", "size": [9.8, 1.5], "position": { "x": -41.887, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-mit-alphafold", "name": "Protein folding solved", "type": "News article", "properties": { "Publisher": "MIT Technology Review", "Article kind": "News report", "Headline": "DeepMind's protein-folding AI has solved a 50-year-old grand challenge of biology", "Date": "2020-11-30", "URL": "https://www.technologyreview.com/2020/11/30/1012712/deepmind-protein-folding-ai-solved-biology-science-drugs-disease/" }, "description": "Will Douglas Heaven of MIT Technology Review reported on 30 November 2020 that DeepMind's AlphaFold had predicted protein shapes in the CASP assessment with an error of about 1.6 angstroms, matching lab methods for the first time. He cited John Moult, who leads the CASP team, calling it the first use of AI to solve a serious problem, and quoted Mohammed AlQuraishi, David Baker and Janet Thornton on the scale of the advance. The piece is enthusiastic but notes that AlphaFold was slower than some rivals and that DeepMind had not yet published how it worked.\n\nhttps://www.technologyreview.com/2020/11/30/1012712/deepmind-protein-folding-ai-solved-biology-science-drugs-disease/", "size": [9.8, 1.5], "position": { "x": -10.2, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-mit-gebru-paper", "name": "Gebru's paper", "type": "News article", "properties": { "Publisher": "MIT Technology Review", "Article kind": "Analysis", "Headline": "We read the paper that forced Timnit Gebru out of Google. Here's what it says.", "Date": "2020-12-04", "URL": "https://www.technologyreview.com/2020/12/04/1013294/google-ai-ethics-research-paper-forced-out-timnit-gebru/" }, "description": "Karen Hao of MIT Technology Review obtained the draft of 'On the Dangers of Stochastic Parrots' from co-author Emily Bender and summarised its four main risks on 4 December 2020: the environmental and financial cost of training, huge and inscrutable training data, the opportunity cost of research, and the illusion of meaning in fluent text. The article quotes Bender, Google's Jeff Dean through his internal email, and Google researcher Nicolas Le Roux, and takes a critical view of Google's account of why the paper was blocked.\n\nhttps://www.technologyreview.com/2020/12/04/1013294/google-ai-ethics-research-paper-forced-out-timnit-gebru/", "size": [9.8, 1.5], "position": { "x": -5.879, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-euronews-ai-act", "name": "Brussels' draft AI rules", "type": "News article", "properties": { "Publisher": "Euronews", "Article kind": "News report", "Headline": "'The higher the risk, the stricter the rule': Brussels' new draft rules on artificial intelligence", "Date": "2021-04-21", "URL": "https://www.euronews.com/my-europe/2021/04/21/the-higher-the-risk-the-stricter-the-rule-brussels-new-draft-rules-on-artificial-intellige" }, "description": "Jorge Liboreiro of Euronews reported on 21 April 2021 on the European Commission's draft AI regulation. The article sets out the four risk tiers, the ban on uses such as government social scoring, the fines of up to 30 million euros or 6% of turnover, and the 1 billion euros a year of planned investment. It quotes Margrethe Vestager and Thierry Breton for the Commission, Sarah Chander of EDRi and Patrick Breyer MEP for the critics, and Cecilia Bonefeld-Dahl of DIGITAL EUROPE for industry. Its angle is a GDPR-style rights framework aimed at competing with the US and China, with biometric surveillance as the most contested issue.\n\nhttps://www.euronews.com/my-europe/2021/04/21/the-higher-the-risk-the-stricter-the-rule-brussels-new-draft-rules-on-artificial-intellige", "size": [9.8, 1.5], "position": { "x": 2.16, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-wapo-lemoine", "name": "'AI came to life'", "type": "News article", "properties": { "Publisher": "The Washington Post", "Article kind": "Feature or interview", "Headline": "The Google engineer who thinks the company's AI has come to life", "Date": "2022-06-11", "URL": "https://www.washingtonpost.com/technology/2022/06/11/google-ai-lamda-blake-lemoine/" }, "description": "Nitasha Tiku of The Washington Post reported on 11 June 2022 on Blake Lemoine, a Google engineer who believed the company's LaMDA chatbot had become sentient and had been placed on paid leave after raising the claim. The story was the first to bring the dispute to a wide audience and set off a wave of coverage and rebuttals.\n\nhttps://www.washingtonpost.com/technology/2022/06/11/google-ai-lamda-blake-lemoine/", "size": [9.8, 1.5], "position": { "x": 26.359, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-nyt-colorado", "name": "AI picture wins art prize", "type": "News article", "properties": { "Publisher": "The New York Times", "Article kind": "News report", "Headline": "An A.I.-Generated Picture Won an Art Prize. Artists Aren't Happy.", "Date": "2022-09-02", "URL": "https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html" }, "description": "Kevin Roose of The New York Times reported on 2 September 2022 on the dispute over Jason Allen's Midjourney image, which had won the digital arts category at the Colorado State Fair. The article was one of the first mainstream reports on artists' objections to image generators.\n\nhttps://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html", "size": [9.8, 1.5], "position": { "x": 33.813, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-nyt-code-red", "name": "Code red for Google search", "type": "News article", "properties": { "Publisher": "The New York Times", "Article kind": "News report", "Headline": "A New Chat Bot Is a 'Code Red' for Google's Search Business", "Date": "2022-12-21", "URL": "https://www.nytimes.com/2022/12/21/technology/ai-chatgpt-google-search.html" }, "description": "The New York Times reported on 21 December 2022 that Google's management had treated ChatGPT as a 'code red' threat to its search business. Reporters Nico Grant and Cade Metz described how executives, including CEO Sundar Pichai, convened to rethink the company's AI strategy and speed up the release of its own conversational products. The article is the origin of the 'code red' framing that later coverage of Google's Bard launch repeated.\n\nhttps://www.nytimes.com/2022/12/21/technology/ai-chatgpt-google-search.html", "size": [9.8, 1.5], "position": { "x": 47.986, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-guardian-pause", "name": "Musk joins pause call", "type": "News article", "properties": { "Publisher": "The Guardian", "Article kind": "News report", "Headline": "Elon Musk joins call for pause in creation of giant AI 'digital minds'", "Date": "2023-03-29", "URL": "https://www.theguardian.com/technology/2023/mar/29/elon-musk-joins-call-for-pause-in-creation-of-giant-ai-digital-minds" }, "description": "The Guardian's Alex Hern reported on 29 March 2023 that Elon Musk had joined a call for a pause in the creation of giant AI 'digital minds', describing the Future of Life Institute letter and its signatories. The Guardian followed up on 1 April 2023 with a piece on the controversy the letter provoked.\n\nhttps://www.theguardian.com/technology/2023/mar/29/elon-musk-joins-call-for-pause-in-creation-of-giant-ai-digital-minds", "size": [9.8, 1.5], "position": { "x": 69.516, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-bloomberg-italy-restore", "name": "ChatGPT back in Italy", "type": "News article", "properties": { "Publisher": "Bloomberg", "Article kind": "News report", "Headline": "OpenAI Restores ChatGPT in Italy, Addresses Regulator Demands", "Date": "2023-04-28", "URL": "https://www.bloomberg.com/news/articles/2023-04-28/openai-restores-chatgpt-in-italy-addresses-regulator-demands" }, "description": "Bloomberg reported on 28 April 2023 that OpenAI had restored ChatGPT access in Italy after meeting conditions set by the data protection authority, ending a ban that began on 31 March. The report summarised the changes OpenAI made, including new disclosures on how training data is used, a form for EU users to object to data use, and an age-verification step at sign-up.\n\nhttps://www.bloomberg.com/news/articles/2023-04-28/openai-restores-chatgpt-in-italy-addresses-regulator-demands", "size": [9.8, 1.5], "position": { "x": 74.564, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-nyt-hinton", "name": "Godfather of AI leaves Google", "type": "News article", "properties": { "Publisher": "The New York Times", "Article kind": "Feature or interview", "Headline": "'The Godfather of A.I.' Leaves Google and Warns of Danger Ahead", "Date": "2023-05-01", "URL": "https://www.nytimes.com/2023/05/01/technology/ai-google-chatbot-engineer-quits-hinton.html" }, "description": "Cade Metz's New York Times interview of 1 May 2023 announced that Geoffrey Hinton had quit Google and explained why: he wanted to speak freely about the risks of the technology he helped create, including misinformation and the possibility of machines surpassing human intelligence. The story carried Hinton's own account of the move, including regret about his life's work, and set off a wave of coverage.\n\nhttps://www.nytimes.com/2023/05/01/technology/ai-google-chatbot-engineer-quits-hinton.html", "size": [9.8, 1.5], "position": { "x": 75.954, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-thr-strike", "name": "Writers Guild strikes", "type": "News article", "properties": { "Publisher": "The Hollywood Reporter", "Article kind": "News report", "Headline": "Writers Guild Calls First Strike in 15 Years", "Date": "2023-05-01", "URL": "https://www.hollywoodreporter.com/business/business-news/wga-writers-strike-moves-forward-contract-negotiations-1235404087/" }, "description": "The Hollywood Reporter's Katie Kilkenny reported on 1 May 2023 that the Writers Guild of America had called its first strike in fifteen years after contract talks with the studios failed. The article set out the guild's demands on pay, staffing and the use of artificial intelligence.\n\nhttps://www.hollywoodreporter.com/business/business-news/wga-writers-strike-moves-forward-contract-negotiations-1235404087/", "size": [9.8, 1.5], "position": { "x": 79.916, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-nvidia-q1-release", "name": "Nvidia Q1 results", "type": "News article", "properties": { "Publisher": "Nvidia Newsroom", "Article kind": "Official release", "Headline": "NVIDIA Announces Financial Results for First Quarter Fiscal 2024", "Date": "2023-05-24", "URL": "https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2024" }, "description": "Nvidia's earnings release of 24 May 2023 reported first-quarter revenue of $7.19 billion, a record $4.28 billion from data centres, and an outlook of $11.00 billion plus or minus 2% for the second quarter. Jensen Huang is quoted saying that the computer industry is going through two simultaneous transitions, accelerated computing and generative AI. The figures sent Nvidia shares up about 24% the next day. It is an official release by the company, not independent reporting.\n\nhttps://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2024", "size": [9.8, 1.5], "position": { "x": 84.964, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-yahoo-nvidia-1t", "name": "Nvidia touches $1 trillion", "type": "News article", "properties": { "Publisher": "Yahoo Finance", "Article kind": "News report", "Headline": "Nvidia touches $1 trillion market cap as chipmaker rides AI wave", "Date": "2023-05-30", "URL": "https://finance.yahoo.com/news/nvidia-touches-1-trillion-market-cap-as-chipmaker-rides-ai-wave-133530381.html" }, "description": "Ines Ferre of Yahoo Finance reported on 30 May 2023 that Nvidia briefly reached a $1 trillion valuation early in the session before dipping back below it, closing at $401.11 and a market value of $992 billion. The article ties the rally to AI demand and the 24 May forecast of $11 billion in second-quarter revenue against Wall Street's $7.2 billion, and quotes Raymond James analyst Srini Pajjuri calling Nvidia the only arms dealer out there in AI.\n\nhttps://finance.yahoo.com/news/nvidia-touches-1-trillion-market-cap-as-chipmaker-rides-ai-wave-133530381.html", "size": [9.8, 1.5], "position": { "x": 90.316, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-verge-extinction", "name": "22-word extinction warning", "type": "News article", "properties": { "Publisher": "The Verge", "Article kind": "News report", "Headline": "Top AI researchers and CEOs warn against 'risk of extinction' in 22-word statement", "Date": "2023-05-30", "URL": "https://www.theverge.com/2023/5/30/23742005/ai-risk-warning-22-word-statement-google-deepmind-openai" }, "description": "The Verge reported on 30 May 2023 on the Center for AI Safety's 22-word statement signed by leading researchers and chief executives from Google DeepMind, OpenAI and Anthropic. The article described who signed and what the statement said, and put it in the context of earlier calls for a pause.\n\nhttps://www.theverge.com/2023/5/30/23742005/ai-risk-warning-22-word-statement-google-deepmind-openai", "size": [9.8, 1.5], "position": { "x": 86.441, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-variety-wga-deal", "name": "WGA deal ends strike", "type": "News article", "properties": { "Publisher": "Variety", "Article kind": "News report", "Headline": "Deal! WGA, AMPTP Reach Historic Contract Agreement to End 146-Day Writers Strike", "Date": "2023-09-24", "URL": "https://variety.com/2023/biz/news/writers-strike-ends-wga-amptp-deal-2-1235733452/" }, "description": "Variety reported on 24 September 2023 that the Writers Guild and the studios' bargaining group had reached a tentative contract that would end the writers' strike after 146 days. The article describes the settlement as historic and covers its main terms, including the new protections against the use of artificial intelligence.\n\nhttps://variety.com/2023/biz/news/writers-strike-ends-wga-amptp-deal-2-1235733452/", "size": [9.8, 1.5], "position": { "x": 104.139, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-guardian-bletchley", "name": "Declaration on AI danger", "type": "News article", "properties": { "Publisher": "The Guardian", "Article kind": "News report", "Headline": "UK, US, EU and China sign declaration of AI's 'catastrophic' danger", "Date": "2023-11-01", "URL": "https://www.theguardian.com/technology/2023/nov/01/uk-us-eu-and-china-sign-declaration-of-ais-catastrophic-danger" }, "description": "The Guardian's Kiran Stacey and Dan Milmo reported on 1 November 2023 that the United Kingdom, the United States, the European Union and China had signed the Bletchley Declaration at the AI Safety Summit, acknowledging the potential for catastrophic harm from advanced AI. A second Guardian piece the same day collected what leaders said at the summit.\n\nhttps://www.theguardian.com/technology/2023/nov/01/uk-us-eu-and-china-sign-declaration-of-ais-catastrophic-danger", "size": [9.8, 1.5], "position": { "x": 112.157, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-business-reporter-277", "name": "Nvidia adds record $277bn", "type": "News article", "properties": { "Publisher": "Reuters", "Article kind": "News report", "Headline": "Nvidia adds record $277 billion in stock market value", "Date": "2024-02-26", "URL": "https://www.business-reporter.co.uk/news/nvidia-adds-record-277-billion-in-stock-market-value-9994" }, "description": "Reuters reporters Noel Randewich and Medha Singh wrote that Nvidia shares rose 16.4% on 22 February 2024 to a record close of $785.38, adding $277 billion in market value, the largest one-day gain in Wall Street history, to reach $1.96 trillion. The report ties the jump to the January-quarter results, which beat expectations on strong demand for AI computing chips, and to a forecast of 233% growth in current-quarter revenue against 208% expected. This copy is republished by Business Reporter on 26 February 2024.\n\nhttps://www.business-reporter.co.uk/news/nvidia-adds-record-277-billion-in-stock-market-value-9994", "size": [9.8, 1.5], "position": { "x": 133.269, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-cnbc-eu-vote", "name": "EU lawmakers pass AI act", "type": "News article", "properties": { "Publisher": "CNBC", "Article kind": "News report", "Headline": "World's first major act to regulate AI passed by European lawmakers", "Date": "2024-03-13", "URL": "https://www.cnbc.com/2024/03/13/european-lawmakers-endorse-worlds-first-major-act-to-regulate-ai.html" }, "description": "CNBC reported on 13 March 2024 that the European Parliament had passed the AI Act, which it described as the world's first major act to regulate artificial intelligence. The report covered the vote, the risk-based approach and the timeline for the rules to take effect.\n\nhttps://www.cnbc.com/2024/03/13/european-lawmakers-endorse-worlds-first-major-act-to-regulate-ai.html", "size": [9.8, 1.5], "position": { "x": 136.994, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-aljazeera-nvidia-3t", "name": "Nvidia dethrones Microsoft", "type": "News article", "properties": { "Publisher": "Al Jazeera", "Article kind": "News report", "Headline": "Nvidia becomes world's most valuable company, dethroning Microsoft", "Date": "2024-06-19", "URL": "https://www.aljazeera.com/economy/2024/6/19/nvidia-becomes-worlds-most-valuable-company-dethroning-microsoft" }, "description": "Al Jazeera reported on 19 June 2024 that Nvidia had passed Microsoft on 18 June to become the world's most valuable listed company, with a market value of $3.335 trillion after its shares rose 3.5% to $135.58. It credits voracious demand for Nvidia's chips from tech giants such as Microsoft, Meta and Google, says the stock was up nearly 182% in 2024 and that Nvidia controls about 80% of the market for AI chips in data centres, and notes that Microsoft fell 0.45% and Apple 1.1%.\n\nhttps://www.aljazeera.com/economy/2024/6/19/nvidia-becomes-worlds-most-valuable-company-dethroning-microsoft", "size": [9.8, 1.5], "position": { "x": 149.576, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-reuters-stargate", "name": "Trump: $500bn AI plan", "type": "News article", "properties": { "Publisher": "Reuters", "Article kind": "News report", "Headline": "OpenAI, SoftBank, Oracle to invest $500 bln in AI, Trump says", "Date": "2025-01-21", "URL": "https://www.reuters.com/technology/artificial-intelligence/openai-softbank-oracle-invest-500-bln-ai-trump-says-2025-01-21/" }, "description": "Reuters reported on 21 January 2025 that OpenAI, SoftBank and Oracle would invest $500 billion in artificial-intelligence infrastructure, as announced by President Trump at the White House. The report set out the joint venture's partners and its first projects.\n\nhttps://www.reuters.com/technology/artificial-intelligence/openai-softbank-oracle-invest-500-bln-ai-trump-says-2025-01-21/", "size": [9.8, 1.5], "position": { "x": 182.164, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-cnbc-deepseek", "name": "DeepSeek tops App Store", "type": "News article", "properties": { "Publisher": "CNBC", "Article kind": "News report", "Headline": "China's DeepSeek AI dethrones ChatGPT on App Store: Here's what you should know", "Date": "2025-01-27", "URL": "https://www.cnbc.com/2025/01/27/chinas-deepseek-ai-tops-chatgpt-app-store-what-you-should-know.html" }, "description": "CNBC reported on 27 January 2025 that DeepSeek's free chatbot app had overtaken ChatGPT as the most-downloaded free app in the US iOS App Store, and explained what the Chinese start-up is. The report was one of the first to connect the app's rise to the day's sell-off in AI-linked stocks.\n\nhttps://www.cnbc.com/2025/01/27/chinas-deepseek-ai-tops-chatgpt-app-store-what-you-should-know.html", "size": [9.8, 1.5], "position": { "x": 179.857, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-nyt-deepseek", "name": "What is DeepSeek?", "type": "News article", "properties": { "Publisher": "The New York Times", "Article kind": "Analysis", "Headline": "What Is DeepSeek? And How Is It Upending A.I.?", "Date": "2025-01-27", "URL": "https://www.nytimes.com/2025/01/27/technology/what-is-deepseek-china-ai.html" }, "description": "Cade Metz of The New York Times published an explainer on 27 January 2025 about DeepSeek, the Chinese start-up whose R1 model and free app had unsettled the US technology industry and financial markets. The article sets out who is behind the company, what it claims to have done with less computing power, and why US rivals and investors were rattled.\n\nhttps://www.nytimes.com/2025/01/27/technology/what-is-deepseek-china-ai.html", "size": [9.8, 1.5], "position": { "x": 179.857, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-bbc-deepseek-shares", "name": "Shares steady after DeepSeek", "type": "News article", "properties": { "Publisher": "BBC News", "Article kind": "News report", "Headline": "Nvidia and Microsoft shares steady after DeepSeek AI app shock", "Date": "2025-01-28", "URL": "https://www.bbc.com/news/articles/c4gpq01rvd4o" }, "description": "BBC News reported on 28 January 2025 that Nvidia and Microsoft shares steadied a day after a sell-off triggered by the arrival of DeepSeek's AI app. According to Wikipedia's article on the AI bubble, which cites the BBC piece, Nvidia had fallen about 17 percent on 27 January, erasing roughly $600 billion in one session, before rebounding about 8.8 percent the next day.\n\nhttps://www.bbc.com/news/articles/c4gpq01rvd4o", "size": [9.8, 1.5], "position": { "x": 190.572, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-hill-rescission", "name": "Biden chip rule rescinded", "type": "News article", "properties": { "Publisher": "The Hill", "Article kind": "News report", "Headline": "Trump administration rescinds Biden AI chip export rule", "Date": "2025-05-13", "URL": "https://www.yahoo.com/news/trump-administration-rescinds-biden-ai-193352703.html" }, "description": "Miranda Nazzaro of The Hill reported on 13 May 2025 that the Commerce Department had rescinded the Biden-era AI diffusion rule, which would have capped AI chip sales to most countries outside 18 allies and partners and was due to take effect that Thursday. It quotes Under Secretary Jeffrey Kessler on a bold, inclusive strategy to spread American AI technology, reports new guidance that using Huawei Ascend chips anywhere violates US export controls, and notes that Microsoft and Nvidia had lobbied against the rule. This copy is syndicated on Yahoo News.\n\nhttps://www.yahoo.com/news/trump-administration-rescinds-biden-ai-193352703.html", "size": [9.8, 1.5], "position": { "x": 209.348, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-decoder-h20", "name": "H20 exports resume", "type": "News article", "properties": { "Publisher": "The Decoder", "Article kind": "News report", "Headline": "Nvidia can resume exports of its H20 AI chip to China after a US policy reversal", "Date": "2025-07-15", "URL": "https://the-decoder.com/nvidia-can-resume-exports-of-its-h20-ai-chip-to-china-after-a-us-policy-reversal/" }, "description": "Maximilian Schreiner of The Decoder wrote on 15 July 2025 that Nvidia would restart H20 sales to China after the US government assured it that licences would be approved, a sharp reversal of the earlier ban. He notes that Jensen Huang had met Donald Trump the week before and was in Beijing for a supply chain expo, that Nvidia had taken multibillion-dollar losses on unsold inventory, and that Huang had called the export controls a failure in May. Nvidia also planned to sell RTX PRO GPUs in China, which it said fall below the thresholds needing special approval.\n\nhttps://the-decoder.com/nvidia-can-resume-exports-of-its-h20-ai-chip-to-china-after-a-us-policy-reversal/", "size": [9.8, 1.5], "position": { "x": 220.836, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-cset-action-plan", "name": "Recapping the AI Action Plan", "type": "News article", "properties": { "Publisher": "Center for Security and Emerging Technology", "Article kind": "Analysis", "Headline": "Trump's Plan for AI: Recapping the White House's AI Action Plan", "Date": "2025-07-23", "URL": "https://cset.georgetown.edu/article/trumps-plan-for-ai-recapping-the-white-houses-ai-action-plan/" }, "description": "The Georgetown Center for Security and Emerging Technology published an analysis on 23 July 2025, the day the White House released its AI Action Plan, summarising its three pillars, innovation, infrastructure and international diplomacy and security, and the three executive orders signed with it on data-centre permitting, exports of the American AI technology stack and 'woke AI' in the federal government. It is a think-tank explainer rather than a news report.\n\nhttps://cset.georgetown.edu/article/trumps-plan-for-ai-recapping-the-white-houses-ai-action-plan/", "size": [9.8, 1.5], "position": { "x": 226.018, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-lemonde-aiact", "name": "Creators condemn 'betrayal'", "type": "News article", "properties": { "Publisher": "Le Monde", "Article kind": "News report", "Headline": "AI Act: 38 global creators' organizations condemn 'betrayal' of Europe's stated goals", "Date": "2025-08-02", "URL": "https://www.lemonde.fr/en/economy/article/2025/08/02/ai-act-38-global-creators-organizations-condemn-betrayal-of-europe-s-stated-goals_6744006_19.html" }, "description": "Le Monde reported on 2 August 2025, the day the EU AI Act's general-purpose AI obligations took effect, that 38 creators' organisations worldwide had condemned the arrangements as a betrayal of Europe's stated goals, arguing that the rules and the accompanying code of practice did not protect authors' and artists' rights over the use of their works to train models.\n\nhttps://www.lemonde.fr/en/economy/article/2025/08/02/ai-act-38-global-creators-organizations-condemn-betrayal-of-europe-s-stated-goals_6744006_19.html", "size": [9.8, 1.5], "position": { "x": 229.176, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-npr-bartz", "name": "Anthropic settles with authors", "type": "News article", "properties": { "Publisher": "NPR", "Article kind": "News report", "Headline": "Anthropic settles with authors in first-of-its-kind AI copyright infringement lawsuit", "Date": "2025-09-05", "URL": "https://www.npr.org/2025/09/05/nx-s1-5529404/anthropic-settlement-authors-copyright-ai" }, "description": "NPR reported on 5 September 2025 that Anthropic had agreed to settle a class-action lawsuit brought by authors who accused it of using pirated books to train its Claude models, in what NPR called a first-of-its-kind settlement in an AI copyright case.\n\nhttps://www.npr.org/2025/09/05/nx-s1-5529404/anthropic-settlement-authors-copyright-ai", "size": [9.8, 1.5], "position": { "x": 236.816, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-tbs-oracle", "name": "Oracle soars on AI cloud", "type": "News article", "properties": { "Publisher": "The Business Standard", "Article kind": "News report", "Headline": "Oracle soars on AI cloud gains, Ellison briefly tops Musk as world's richest man", "Date": "2025-09-11", "URL": "https://www.tbsnews.net/node/1233721" }, "description": "A Reuters report republished by The Business Standard on 11 September 2025 described how Oracle shares jumped about 43 percent at their peak on 10 September and were up 36.7 percent late in the session, adding about $234 billion of market value, after Oracle disclosed four multi-billion-dollar contracts and the Wall Street Journal reported a $300 billion computing deal with OpenAI. It adds that Larry Ellison's fortune rose by about $100 billion to roughly $392.6 billion, according to Forbes, briefly putting him above Elon Musk.\n\nhttps://www.tbsnews.net/node/1233721", "size": [9.8, 1.5], "position": { "x": 241.81, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-entrepreneur-nvidia-5t", "name": "First $5 trillion company", "type": "News article", "properties": { "Publisher": "Entrepreneur", "Article kind": "News report", "Headline": "Nvidia Becomes the First Company to Reach a $5 Trillion Market Value, Leaving Apple and Microsoft in the Dust", "Date": "2025-10-29", "URL": "https://www.entrepreneur.com/business-news/nvidia-becomes-the-first-to-pass-5-trillion-in-market-value" }, "description": "Sherin Shibu of Entrepreneur reported on 29 October 2025 that Nvidia had become the first company worth $5 trillion, ahead of Apple and Microsoft at about $4 trillion each, with a share price above $210, up over 4% in the morning and more than 50% for the year. It credits Jensen Huang's announcement of about $500 billion in expected AI chip orders and seven US government supercomputers, and quotes Truist's Keith Lerner saying a $5 trillion market cap would have been unimaginable a few years ago and that the market is putting a lot of stock into AI being transformational.\n\nhttps://www.entrepreneur.com/business-news/nvidia-becomes-the-first-to-pass-5-trillion-in-market-value", "size": [9.8, 1.5], "position": { "x": 250.773, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-fortune-pentagon", "name": "Supply-chain risk", "type": "News article", "properties": { "Publisher": "Fortune", "Article kind": "News report", "Headline": "Pentagon officially defines Anthropic as 'supply chain risk'", "Date": "2026-03-06", "URL": "https://fortune.com/2026/03/06/pentagon-officially-defines-anthropic-as-supply-chain-risk/" }, "description": "An Associated Press report published by Fortune on 6 March 2026 described how the Pentagon formally told Anthropic that it and its products are a supply chain risk after a week of threats from President Trump and Defense Secretary Pete Hegseth. The report covers Anthropic's refusal to remove limits on mass surveillance and autonomous weapons, Lockheed Martin's and Microsoft's responses, Senator Gillibrand's criticism, Dario Amodei's pledge to sue, OpenAI's replacement deal, and a surge of consumer sign-ups for Claude, with more than a million a day and a top App Store ranking in over twenty countries.\n\nhttps://fortune.com/2026/03/06/pentagon-officially-defines-anthropic-as-supply-chain-risk/", "size": [9.8, 1.5], "position": { "x": 274.031, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-cbs-sora", "name": "OpenAI pulls plug on Sora", "type": "News article", "properties": { "Publisher": "CBS News", "Article kind": "News report", "Headline": "OpenAI pulls the plug on its Sora AI video app", "Date": "2026-03-24", "URL": "https://www.cbsnews.com/news/sora-ai-openai-discontinues/" }, "description": "CBS News reported on 24 March 2026 that OpenAI was shutting down its Sora AI video app and API, quoting the company's statement that it had decided to discontinue Sora and that it was weighing how to allocate compute. The report notes that the Wall Street Journal first reported the decision and described the platform as waning in popularity, that ChatGPT's image generator would remain, and that Disney's December 2025 licensing deal covering hundreds of characters would be affected.\n\nhttps://www.cbsnews.com/news/sora-ai-openai-discontinues/", "size": [9.8, 1.5], "position": { "x": 276.839, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-wsj-musk-verdict", "name": "Jury rejects Musk's claims", "type": "News article", "properties": { "Publisher": "The Wall Street Journal", "Article kind": "News report", "Headline": "Jury Rejects Musk's Claims Against OpenAI", "Date": "2026-05-18", "URL": "https://www.wsj.com/tech/ai/jury-sides-with-openai-sam-altman-in-case-brought-by-elon-musk-933240ff" }, "description": "The Wall Street Journal reported on 18 May 2026 that a federal jury had rejected Elon Musk's claims against OpenAI and Sam Altman, finding them barred by the statute of limitations. The report closed the story of a three-week trial over OpenAI's turn from a non-profit lab to a for-profit company, in which Musk and Microsoft chief executive Satya Nadella had testified.\n\nhttps://www.wsj.com/tech/ai/jury-sides-with-openai-sam-altman-in-case-brought-by-elon-musk-933240ff", "size": [9.8, 1.5], "position": { "x": 286.463, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-gibsondunn-omnibus", "name": "AI Act omnibus deal", "type": "News article", "properties": { "Publisher": "Gibson Dunn", "Article kind": "Analysis", "Headline": "EU AI Act Omnibus Agreement — Postponed High-Risk Deadlines and Other Key Changes", "Date": "2026-05-27", "URL": "https://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines/" }, "description": "A client alert by the law firm Gibson Dunn dated 27 May 2026 explains the provisional agreement on the EU Digital Omnibus on AI: stand-alone high-risk obligations postponed to 2 December 2027 and embedded-product obligations to 2 August 2028, a new prohibition on non-consensual intimate imagery and child sexual abuse material, a watermarking grace period to 2 December 2026, wider AI Office powers and softened AI-literacy duties. It states that the political agreement was reached on 6 May 2026 and that formal adoption and publication were expected before 2 August 2026. It is a legal-industry analysis rather than a news report.\n\nhttps://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines/", "size": [9.8, 1.5], "position": { "x": 289.527, "y": -7.19, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-boom-na-aljazeera-anthropic-lifted", "name": "US lifts Anthropic curbs", "type": "News article", "properties": { "Publisher": "Al Jazeera", "Article kind": "News report", "Headline": "US lifts restrictions on Anthropic’s powerful AI models Fable and Mythos", "Date": "2026-07-01", "URL": "https://www.aljazeera.com/economy/2026/7/1/us-lifts-restrictions-on-powerful-ai-models-fable-mythos-anthropic-says" }, "description": "Al Jazeera reported on 1 July 2026 that the US Department of Commerce had lifted its export controls on Anthropic's Claude Fable 5 and Mythos 5, which Anthropic had shut off abruptly the previous month after being ordered to block all foreign nationals. The article quotes the terms of a letter from Commerce Secretary Howard Lutnick under which Anthropic would no longer need an export licence and in return agreed to detect security risks proactively, work with the government on standards and report malicious activity, and notes that officials gave no specific reason for the restrictions.\n\nhttps://www.aljazeera.com/economy/2026/7/1/us-lifts-restrictions-on-powerful-ai-models-fable-mythos-anthropic-says", "size": [9.8, 1.5], "position": { "x": 297.261, "y": -9.29, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "ai-scale-na-fnnews-h200", "name": "Few H200 chips reach China", "type": "News article", "properties": { "Publisher": "Fnnews", "Article kind": "News report", "Headline": "U.S. Commerce Department Says Only Tiny Number of NVIDIA H200 Chips Have Been Exported to China", "Date": "2026-07-15", "URL": "https://en.fnnews.com/news/202607150755449382" }, "description": "The Korean outlet Fnnews reported on 15 July 2026 that Jeffrey Kessler, Under Secretary of Commerce for Industry and Security, told a congressional hearing that actual shipments under export licences for the H200 and comparable chips had been extremely rare and that only a very small number had reached China. The article calls it the first official confirmation that exports had resumed, notes that the government reviews Chinese applicants case by case with on-site inspections, that some licences were denied, and that Nvidia declined to comment.\n\nhttps://en.fnnews.com/news/202607150755449382", "size": [9.8, 1.5], "position": { "x": 299.691, "y": -5.09, "z": 0 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-band-M2", "name": "People", "type": "the grid", "properties": {}, "size": [15, 17.86], "description": "", "position": { "x": -307.465, "y": 62.89, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-band-M1", "name": "Organisations", "type": "the grid", "properties": {}, "size": [15, 6], "description": "", "position": { "x": -307.465, "y": 49.76, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-band-S1", "name": "Ideas (papers)", "type": "the grid", "properties": {}, "size": [15, 6.44], "description": "", "position": { "x": -307.465, "y": 42.34, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-band-S2", "name": "Systems and products", "type": "the grid", "properties": {}, "size": [15, 33.12], "description": "", "position": { "x": -307.465, "y": 21.36, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-band-S3", "name": "Compute", "type": "the grid", "properties": {}, "size": [15, 6.74], "description": "", "position": { "x": -307.465, "y": 0.23, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-band-C1", "name": "In the press", "type": "the grid", "properties": {}, "size": [15, 5.7], "description": "", "position": { "x": -307.465, "y": -7.19, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-band-C2", "name": "Reactions", "type": "the grid", "properties": {}, "size": [15, 19.94], "description": "", "position": { "x": -307.465, "y": -21.21, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-band-C3", "name": "Rules", "type": "the grid", "properties": {}, "size": [15, 21.24], "description": "", "position": { "x": -307.465, "y": -43, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-band-C4", "name": "Money", "type": "the grid", "properties": {}, "size": [15, 24.4], "description": "", "position": { "x": -307.465, "y": -67.02, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-1950s", "name": "1950s", "type": "the grid", "properties": {}, "size": [15.51, 8], "description": "", "position": { "x": -278.71, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-1960s", "name": "1960s", "type": "the grid", "properties": {}, "size": [17.309, 8], "description": "", "position": { "x": -261.701, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-1970s", "name": "1970s", "type": "the grid", "properties": {}, "size": [18.52, 8], "description": "", "position": { "x": -243.186, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-1980s", "name": "1980s", "type": "the grid", "properties": {}, "size": [24.104, 8], "description": "", "position": { "x": -221.274, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-1990s", "name": "1990s", "type": "the grid", "properties": {}, "size": [17.344, 8], "description": "", "position": { "x": -199.95, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2000s", "name": "2000s", "type": "the grid", "properties": {}, "size": [24.095, 8], "description": "", "position": { "x": -178.631, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2010", "name": "2010", "type": "the grid", "properties": {}, "size": [3.129, 8], "description": "", "position": { "x": -164.419, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2011", "name": "2011", "type": "the grid", "properties": {}, "size": [7.189, 8], "description": "", "position": { "x": -158.659, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2012", "name": "2012", "type": "the grid", "properties": {}, "size": [13.274, 8], "description": "", "position": { "x": -147.828, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2013", "name": "2013", "type": "the grid", "properties": {}, "size": [10.679, 8], "description": "", "position": { "x": -135.251, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2014", "name": "2014", "type": "the grid", "properties": {}, "size": [14.935, 8], "description": "", "position": { "x": -121.844, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2015", "name": "2015", "type": "the grid", "properties": {}, "size": [13.532, 8], "description": "", "position": { "x": -107.011, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2016", "name": "2016", "type": "the grid", "properties": {}, "size": [22.579, 8], "description": "", "position": { "x": -88.355, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2017", "name": "2017", "type": "the grid", "properties": {}, "size": [17.589, 8], "description": "", "position": { "x": -67.671, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2018", "name": "2018", "type": "the grid", "properties": {}, "size": [12.204, 8], "description": "", "position": { "x": -52.174, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2019", "name": "2019", "type": "the grid", "properties": {}, "size": [17.031, 8], "description": "", "position": { "x": -36.956, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2020", "name": "2020", "type": "the grid", "properties": {}, "size": [23.36, 8], "description": "", "position": { "x": -16.161, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2021", "name": "2021", "type": "the grid", "properties": {}, "size": [19.5, 8], "description": "", "position": { "x": 5.869, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2022", "name": "2022", "type": "the grid", "properties": {}, "size": [34.099, 8], "description": "", "position": { "x": 33.268, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2023", "name": "2023", "type": "the grid", "properties": {}, "size": [73.569, 8], "description": "", "position": { "x": 87.702, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2024", "name": "2024", "type": "the grid", "properties": {}, "size": [50.752, 8], "description": "", "position": { "x": 150.463, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2025", "name": "2025", "type": "the grid", "properties": {}, "size": [86.426, 8], "description": "", "position": { "x": 219.652, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } }, { "ID": "kl-tick-2026", "name": "2026", "type": "the grid", "properties": {}, "size": [48.701, 8], "description": "", "position": { "x": 287.815, "y": 80.52, "z": 0.01 }, "features": { "resizeTextToFit": true } } ], "links": [ { "node1": "ai-p-joseph-weizenbaum", "node2": "ai-hist-ms-eliza", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "wrote ELIZA at MIT 1964 to 1967" } }, { "node1": "ai-o-mit", "node2": "ai-hist-ms-eliza", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "ELIZA was built at MIT" } }, { "node1": "ai-p-joseph-weizenbaum", "node2": "ai-o-mit", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "MIT professor from 1963" } }, { "node1": "ai-hist-ms-eliza", "node2": "ai-hist-re-weizenbaum-critique", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the reception of ELIZA led Weizenbaum to write his 1976 critique" } }, { "node1": "ai-p-joseph-weizenbaum", "node2": "ai-hist-re-weizenbaum-critique", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "author of Computer Power and Human Reason" } }, { "node1": "ai-hist-o-sri", "node2": "ai-hist-ms-shakey", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "built at SRI's Artificial Intelligence Center 1966 to 1972" } }, { "node1": "ai-o-darpa", "node2": "ai-hist-ms-shakey", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "DARPA funded the Shakey project" } }, { "node1": "ai-p-james-lighthill", "node2": "ai-a-lighthill-report", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "author of the 1973 report for the Science Research Council" } }, { "node1": "ai-a-lighthill-report", "node2": "ai-era-3", "name1": "is opened by", "name2": "opens", "direction": 1, "properties": { "Relation": "causation", "Link kind": "era boundary", "Reason": "the 1973 report led the UK to end most university AI support and marks the start of the first winter" } }, { "node1": "ai-hist-pa-perceptrons-book", "node2": "ai-a-lighthill-report", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the 1969 book and the Lighthill report are cited together as causes of the first winter" } }, { "node1": "ai-era-2", "node2": "ai-era-3", "name1": "follows", "name2": "precedes", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "era boundary", "Reason": "Early optimism ends in 1973 and the first winter begins in 1974" } }, { "node1": "ai-o-stanford", "node2": "ai-hist-ms-mycin", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Shortliffe built MYCIN at Stanford as doctoral work" } }, { "node1": "ai-p-john-mccarthy", "node2": "ai-o-stanford", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "professor at Stanford 1962 to 2000" } }, { "node1": "ai-hist-o-carnegie-mellon", "node2": "ai-hist-ms-xcon", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "John McDermott of Carnegie Mellon wrote XCON for DEC" } }, { "node1": "ai-hist-ms-mycin", "node2": "ai-hist-ms-xcon", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "both are rule-based expert systems; XCON followed MYCIN's rule-based approach" } }, { "node1": "ai-hist-ms-xcon", "node2": "ai-era-4", "name1": "is opened by", "name2": "opens", "direction": 1, "properties": { "Relation": "causation", "Link kind": "era boundary", "Reason": "first commercial use in 1980 marks the expert systems boom" } }, { "node1": "ai-era-3", "node2": "ai-era-4", "name1": "follows", "name2": "precedes", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "era boundary", "Reason": "first winter ends 1979, expert systems boom begins 1980" } }, { "node1": "ai-a-fifth-generation-project", "node2": "ai-a-strategic-computing", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the US launched the Strategic Computing Initiative in response to Japan's project" } }, { "node1": "ai-o-darpa", "node2": "ai-a-strategic-computing", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "DARPA funded and ran the programme" } }, { "node1": "ai-a-strategic-computing", "node2": "ai-hist-ev-lisp-machine-collapse", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the 1987 to 1988 DARPA cuts and the Lisp machine collapse are given together as the start of the second winter" } }, { "node1": "ai-hist-ev-lisp-machine-collapse", "node2": "ai-era-5", "name1": "is opened by", "name2": "opens", "direction": 1, "properties": { "Relation": "causation", "Link kind": "era boundary", "Reason": "1987 collapse starts the second AI winter" } }, { "node1": "ai-era-4", "node2": "ai-era-5", "name1": "follows", "name2": "precedes", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "era boundary", "Reason": "expert systems boom ends 1987, second winter starts 1988" } }, { "node1": "ai-p-john-hopfield", "node2": "ai-hist-pa-hopfield-1982", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "sole author of the 1982 PNAS paper" } }, { "node1": "ai-p-john-hopfield", "node2": "ai-a-nobel-physics-2024", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "recognition", "Reason": "shared the 2024 Nobel Prize in Physics with Hinton" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-a-nobel-physics-2024", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "recognition", "Reason": "shared the 2024 Nobel Prize in Physics with Hopfield" } }, { "node1": "ai-hist-pa-hopfield-1982", "node2": "ai-a-nobel-physics-2024", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "recognition", "Reason": "the Nobel citation names the Hopfield network among the foundational discoveries" } }, { "node1": "ai-hist-p-david-rumelhart", "node2": "ai-a-backprop-1986", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first author of the 1986 Nature paper" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-a-backprop-1986", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author of the 1986 Nature paper" } }, { "node1": "ai-a-backprop-1986", "node2": "ai-hist-ms-lenet", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "LeCun's group applied backpropagation to handwritten zip code recognition" } }, { "node1": "ai-p-yann-lecun", "node2": "ai-hist-ms-lenet", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "led the Bell Labs group that developed LeNet" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-p-yann-lecun", "name1": "is supervised by", "name2": "supervises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "mentorship", "Reason": "LeCun was a postdoctoral researcher under Hinton in Toronto" } }, { "node1": "ai-hist-p-sepp-hochreiter", "node2": "ai-a-lstm", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first author of the 1997 LSTM paper" } }, { "node1": "ai-p-juergen-schmidhuber", "node2": "ai-a-lstm", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author of the 1997 LSTM paper" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-a-turing-award-2018", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "recognition", "Reason": "shared the 2018 Turing Award with Bengio and LeCun" } }, { "node1": "ai-p-yann-lecun", "node2": "ai-a-turing-award-2018", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "recognition", "Reason": "shared the 2018 Turing Award with Hinton and Bengio" } }, { "node1": "ai-p-yoshua-bengio", "node2": "ai-a-turing-award-2018", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "recognition", "Reason": "shared the 2018 Turing Award with Hinton and LeCun" } }, { "node1": "ai-o-ibm", "node2": "ai-a-deep-blue", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "IBM built the Deep Blue chess computer" } }, { "node1": "ai-a-deep-blue", "node2": "ai-p-garry-kasparov", "name1": "is beaten by", "name2": "beats", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "won the 1997 rematch 3.5 to 2.5" } }, { "node1": "ai-a-deep-blue", "node2": "ai-hist-ev-deep-blue-match-1997", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "the machine won the six-game match" } }, { "node1": "ai-hist-na-chessbase-cheating-move-2015", "node2": "ai-a-deep-blue", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "analysis of a Deep Blue move" } }, { "node1": "ai-o-ibm", "node2": "ai-hist-ms-watson", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "IBM's DeepQA project built Watson" } }, { "node1": "ai-hist-ms-watson", "node2": "ai-hist-ev-jeopardy-2011", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "beat Jennings and Rutter in February 2011" } }, { "node1": "ai-o-darpa", "node2": "ai-hist-ev-darpa-grand-challenge-2005", "name1": "is organised by", "name2": "organises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "DARPA ran the Grand Challenge and paid the 2 million dollar prize" } }, { "node1": "ai-hist-ms-stanley", "node2": "ai-hist-ev-darpa-grand-challenge-2005", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "first of five finishers, winner of the prize" } }, { "node1": "ai-o-stanford", "node2": "ai-hist-ms-stanley", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "built by the Stanford Racing Team" } }, { "node1": "ai-hist-p-sebastian-thrun", "node2": "ai-hist-ms-stanley", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "led the Stanford Racing Team" } }, { "node1": "ai-hist-p-sebastian-thrun", "node2": "ai-o-stanford", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "joined Stanford in 2003 and re-established its AI lab" } }, { "node1": "ai-p-fei-fei-li", "node2": "ai-a-imagenet", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "led the ImageNet project and co-authored the 2009 paper" } }, { "node1": "ai-p-fei-fei-li", "node2": "ai-o-stanford", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "Stanford professor since 2009" } }, { "node1": "ai-a-alexnet", "node2": "ai-a-imagenet", "name1": "is the benchmark for", "name2": "is evaluated on", "direction": 1, "properties": { "Relation": "information", "Link kind": "benchmark", "Reason": "AlexNet won the ImageNet challenge in 2012" } }, { "node1": "ai-a-imagenet", "node2": "ai-hist-ev-ilsvrc-2012", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "benchmark", "Reason": "the challenge uses a 1,000-class subset of ImageNet" } }, { "node1": "ai-a-alexnet", "node2": "ai-hist-ev-ilsvrc-2012", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "SuperVision had the best top-5 error of 15.3 percent" } }, { "node1": "ai-a-alexnet", "node2": "ai-era-6", "name1": "is opened by", "name2": "opens", "direction": 1, "properties": { "Relation": "causation", "Link kind": "era boundary", "Reason": "the 2012 ImageNet result starts the deep learning revolution" } }, { "node1": "ai-era-5", "node2": "ai-era-6", "name1": "follows", "name2": "precedes", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "era boundary", "Reason": "the second winter and statistical learning ends the day before the AlexNet result" } }, { "node1": "ai-p-alex-krizhevsky", "node2": "ai-a-alexnet", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "lead author who built the network" } }, { "node1": "ai-p-ilya-sutskever", "node2": "ai-a-alexnet", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "co-developer of AlexNet" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-a-alexnet", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "supervisor and co-author of the AlexNet paper" } }, { "node1": "ai-o-university-of-toronto", "node2": "ai-a-alexnet", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "AlexNet was built at the University of Toronto" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-o-university-of-toronto", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "professor at Toronto since 1987" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-p-ilya-sutskever", "name1": "is supervised by", "name2": "supervises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "mentorship", "Reason": "Sutskever earned his Toronto PhD under Hinton" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-p-alex-krizhevsky", "name1": "is supervised by", "name2": "supervises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "mentorship", "Reason": "Krizhevsky earned his Toronto PhD under Hinton" } }, { "node1": "ai-hist-ms-lenet", "node2": "ai-a-alexnet", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "AlexNet is a deeper convolutional network in the lineage of LeNet" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-hist-o-dnnresearch", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "founded DNNresearch in 2012 with his students" } }, { "node1": "ai-p-alex-krizhevsky", "node2": "ai-hist-o-dnnresearch", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founder of DNNresearch" } }, { "node1": "ai-p-ilya-sutskever", "node2": "ai-hist-o-dnnresearch", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founder of DNNresearch" } }, { "node1": "ai-a-alexnet", "node2": "ai-hist-fr-google-dnnresearch", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the authors sold their company to Google after the 2012 result" } }, { "node1": "ai-o-google", "node2": "ai-hist-fr-google-dnnresearch", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Google was the acquirer" } }, { "node1": "ai-hist-fr-google-dnnresearch", "node2": "ai-hist-o-dnnresearch", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "the company that was bought" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-o-google", "name1": "is joined by", "name2": "leaves for", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "move", "Reason": "joined Google after the DNNresearch purchase and worked there until May 2023" } }, { "node1": "ai-p-andrew-ng", "node2": "ai-hist-pa-gpu-deep-learning-2009", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "third author of the 2009 GPU paper" } }, { "node1": "ai-hist-pa-deep-belief-nets", "node2": "ai-hist-pa-gpu-deep-learning-2009", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "the GPU paper accelerates the training of deep belief networks" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-hist-pa-deep-belief-nets", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first author of the 2006 paper" } }, { "node1": "ai-hist-pa-gpu-deep-learning-2009", "node2": "ai-a-alexnet", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "showed GPUs could train large networks; AlexNet was trained on two GPUs" } }, { "node1": "ai-a-alexnet", "node2": "ai-a-gtx-580", "name1": "is used to train", "name2": "is trained on", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "training compute", "Reason": "trained on two Nvidia GTX 580 GPUs" } }, { "node1": "ai-a-alexnet", "node2": "ai-a-resnet", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "AlexNet's architecture shaped later networks including ResNet" } }, { "node1": "ai-a-backprop-1986", "node2": "ai-a-alexnet", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "AlexNet is trained with backpropagation" } }, { "node1": "ai-o-google", "node2": "ai-hist-ms-google-cat-network", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "built by the Google Brain team" } }, { "node1": "ai-p-andrew-ng", "node2": "ai-hist-ms-google-cat-network", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "co-founder of Google Brain and last author of the paper" } }, { "node1": "ai-p-andrew-ng", "node2": "ai-o-google", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "founded the Google Brain project in 2011" } }, { "node1": "ai-p-fei-fei-li", "node2": "ai-hist-ev-ilsvrc-2012", "name1": "is organised by", "name2": "organises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "creator of ImageNet, the data behind the challenge" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-hist-pa-dropout", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first author of the 2012 dropout preprint" } }, { "node1": "ai-p-alex-krizhevsky", "node2": "ai-hist-pa-dropout", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author of the dropout preprint" } }, { "node1": "ai-p-ilya-sutskever", "node2": "ai-hist-pa-dropout", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author of the dropout preprint" } }, { "node1": "ai-hist-pa-dropout", "node2": "ai-a-alexnet", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "the AlexNet paper reports using a new regularization method against overfitting, the same authors' dropout" } }, { "node1": "ai-hist-p-tomas-mikolov", "node2": "ai-hist-pa-word2vec", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "lead author" } }, { "node1": "ai-p-ilya-sutskever", "node2": "ai-hist-pa-word2vec", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author of the word2vec papers" } }, { "node1": "ai-o-google", "node2": "ai-hist-pa-word2vec", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "written by Mikolov and colleagues at Google" } }, { "node1": "ai-hist-pa-word2vec", "node2": "ai-a-bert", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "contextual models such as ELMo and BERT superseded static word embeddings" } }, { "node1": "ai-p-ian-goodfellow", "node2": "ai-a-gans", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first author" } }, { "node1": "ai-p-yoshua-bengio", "node2": "ai-a-gans", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "last author of the GAN paper" } }, { "node1": "ai-p-yoshua-bengio", "node2": "ai-p-ian-goodfellow", "name1": "is supervised by", "name2": "supervises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "mentorship", "Reason": "Bengio advised Goodfellow's Montreal PhD" } }, { "node1": "ai-p-yoshua-bengio", "node2": "ai-hist-pa-rnn-encoder-decoder", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "last author of the paper" } }, { "node1": "ai-hist-pa-rnn-encoder-decoder", "node2": "ai-a-bahdanau-attention", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "the attention paper starts from the encoder-decoder approach and shares authors" } }, { "node1": "ai-p-yoshua-bengio", "node2": "ai-a-bahdanau-attention", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author with Bahdanau and Cho" } }, { "node1": "ai-a-bahdanau-attention", "node2": "ai-a-transformer", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "attention for translation led to self-attention in the Transformer" } }, { "node1": "ai-a-bahdanau-attention", "node2": "ai-hist-ms-gnmt", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "GNMT uses attention" } }, { "node1": "ai-p-ilya-sutskever", "node2": "ai-hist-pa-seq2seq", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first author" } }, { "node1": "ai-o-google", "node2": "ai-hist-pa-seq2seq", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "written at Google Brain" } }, { "node1": "ai-a-lstm", "node2": "ai-hist-pa-seq2seq", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "encoder and decoder are both LSTMs" } }, { "node1": "ai-hist-pa-seq2seq", "node2": "ai-hist-ms-gnmt", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "the research helped move Google Translate to neural machine translation" } }, { "node1": "ai-hist-pa-seq2seq", "node2": "ai-a-transformer", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "the Transformer replaced the recurrent encoder and decoder of seq2seq with self-attention" } }, { "node1": "ai-o-google", "node2": "ai-hist-ms-gnmt", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Google's production translation system" } }, { "node1": "ai-p-yann-lecun", "node2": "ai-hist-pa-nature-deep-learning", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first author of the Nature review" } }, { "node1": "ai-p-yoshua-bengio", "node2": "ai-hist-pa-nature-deep-learning", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author of the Nature review" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-hist-pa-nature-deep-learning", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author of the Nature review" } }, { "node1": "ai-hist-p-kaiming-he", "node2": "ai-a-resnet", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first author" } }, { "node1": "ai-o-microsoft", "node2": "ai-a-resnet", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "He worked at Microsoft Research Asia from 2011 to 2016" } }, { "node1": "ai-a-resnet", "node2": "ai-hist-ms-gnmt", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "GNMT uses residual connections" } }, { "node1": "ai-o-deepmind", "node2": "ai-hist-ms-dqn", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "the Atari-playing Deep Q-network was DeepMind's work" } }, { "node1": "ai-hist-p-david-silver", "node2": "ai-hist-ms-dqn", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "co-author of the DQN papers" } }, { "node1": "ai-hist-ms-dqn", "node2": "ai-hist-ms-alphago", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "AlphaGo's networks were refined by reinforcement learning developed in the DQN line" } }, { "node1": "ai-o-deepmind", "node2": "ai-hist-ms-alphago", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "AlphaGo was developed by Google DeepMind" } }, { "node1": "ai-hist-p-david-silver", "node2": "ai-hist-ms-alphago", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "led the AlphaGo project" } }, { "node1": "ai-p-demis-hassabis", "node2": "ai-hist-ms-alphago", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "co-author of the AlphaGo paper and announcement" } }, { "node1": "ai-hist-ms-alphago", "node2": "ai-p-lee-sedol", "name1": "is beaten by", "name2": "beats", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "won 4 to 1 in March 2016" } }, { "node1": "ai-hist-ms-alphago", "node2": "ai-a-alphago-lee-sedol", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "won the match 4 to 1" } }, { "node1": "ai-a-alphago-lee-sedol", "node2": "ai-hist-re-alphago-shock-korea", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the defeat of Lee Sedol was received as a shock in South Korea" } }, { "node1": "ai-a-alphago-lee-sedol", "node2": "ai-hist-re-alphago-china", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the match set off a debate in China" } }, { "node1": "ai-hist-re-alphago-shock-korea", "node2": "ai-hist-fr-south-korea-ai-fund", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the government announced the fund as a response to the AlphaGo shock" } }, { "node1": "ai-a-alphago-lee-sedol", "node2": "ai-hist-fr-south-korea-ai-fund", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "fund announced two days after the match ended" } }, { "node1": "ai-hist-re-alphago-china", "node2": "ai-a-china-ai-plan-2017", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the AlphaGo match is widely cited as a spur for China's 2017 AI plan" } }, { "node1": "ai-hist-na-koreatimes-lee-beatable-2016", "node2": "ai-a-alphago-lee-sedol", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "report on Lee's win in Game 4" } }, { "node1": "ai-hist-na-koreatimes-lee-beatable-2016", "node2": "ai-p-lee-sedol", "name1": "is quoted in", "name2": "quotes", "direction": 1, "properties": { "Relation": "information", "Link kind": "quotation", "Reason": "quotes Lee calling the win priceless" } }, { "node1": "ai-hist-na-koreatimes-lee-beatable-2016", "node2": "ai-p-demis-hassabis", "name1": "is quoted in", "name2": "quotes", "direction": 1, "properties": { "Relation": "information", "Link kind": "quotation", "Reason": "quotes Hassabis saying Lee was too strong that day" } }, { "node1": "ai-hist-na-joongang-alphago-shock-2016", "node2": "ai-hist-re-alphago-shock-korea", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "the editorial is itself an instance and account of the reaction" } }, { "node1": "ai-hist-na-foreignpolicy-alphago-2016", "node2": "ai-hist-re-alphago-china", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "report on the Chinese debate after the match" } }, { "node1": "ai-hist-na-foreignpolicy-alphago-2016", "node2": "ai-p-lee-sedol", "name1": "is quoted in", "name2": "quotes", "direction": 1, "properties": { "Relation": "information", "Link kind": "quotation", "Reason": "quotes Lee's postgame comment about Ke Jie" } }, { "node1": "ai-p-demis-hassabis", "node2": "ai-o-deepmind", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founded DeepMind in 2010" } }, { "node1": "ai-p-mustafa-suleyman", "node2": "ai-o-deepmind", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founded DeepMind in 2010" } }, { "node1": "ai-o-google", "node2": "ai-a-deepmind-acquisition", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Google was the acquirer" } }, { "node1": "ai-a-deepmind-acquisition", "node2": "ai-o-deepmind", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "DeepMind was the company bought" } }, { "node1": "ai-p-sam-altman", "node2": "ai-o-openai", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founder and co-chair at launch" } }, { "node1": "ai-p-elon-musk", "node2": "ai-o-openai", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founder and co-chair at launch" } }, { "node1": "ai-hist-p-greg-brockman", "node2": "ai-o-openai", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founder" } }, { "node1": "ai-p-ilya-sutskever", "node2": "ai-o-openai", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founder and research director at launch" } }, { "node1": "ai-o-amazon", "node2": "ai-hist-fr-openai-founding", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Amazon Web Services was among the named funders" } }, { "node1": "ai-hist-fr-openai-founding", "node2": "ai-o-openai", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "the pledge funded the new lab" } }, { "node1": "ai-o-apple", "node2": "ai-hist-pr-siri", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Apple announced and shipped Siri with the iPhone 4S" } }, { "node1": "ai-hist-o-sri", "node2": "ai-hist-pr-siri", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Siri spun out of SRI's DARPA-funded CALO project" } }, { "node1": "ai-p-yann-lecun", "node2": "ai-o-meta", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "first director of Facebook AI Research from 2013" } }, { "node1": "ai-hist-p-tomas-mikolov", "node2": "ai-o-meta", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "Facebook AI Research 2014 to 2020" } }, { "node1": "ai-hist-p-kaiming-he", "node2": "ai-o-meta", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "Facebook AI Research 2016 to 2024" } }, { "node1": "ai-p-elon-musk", "node2": "ai-hist-re-fli-open-letter-2015", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "among the signatories of the January 2015 letter" } }, { "node1": "ai-o-future-of-life", "node2": "ai-hist-re-fli-open-letter-2015", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "the Future of Life Institute drafted and published the letter" } }, { "node1": "ai-o-white-house", "node2": "ai-hist-rg-us-ai-rd-strategic-plan", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "issued through the National Science and Technology Council in the Executive Office of the President" } }, { "node1": "ai-hist-ms-google-cat-network", "node2": "ai-hist-pa-word2vec", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "both came from the Google Brain group; the cat network showed large-scale unsupervised learning working" } }, { "node1": "ai-hist-ev-ilsvrc-2012", "node2": "ai-a-resnet", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "ResNet won the ILSVRC 2015 classification task, continuing the ImageNet contest" } }, { "node1": "ai-p-alan-turing", "node2": "ai-a-turing-1950", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "sole author of the 1950 paper in Mind" } }, { "node1": "ai-a-turing-1950", "node2": "ai-era-1", "name1": "is opened by", "name2": "opens", "direction": 1, "properties": { "Relation": "causation", "Link kind": "era boundary", "Reason": "the 1950 paper is the conventional start of the history of AI" } }, { "node1": "ai-era-1", "node2": "ai-era-2", "name1": "follows", "name2": "precedes", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "era boundary", "Reason": "Foundations ends where the Dartmouth workshop begins" } }, { "node1": "ai-a-dartmouth-workshop", "node2": "ai-era-2", "name1": "is opened by", "name2": "opens", "direction": 1, "properties": { "Relation": "causation", "Link kind": "era boundary", "Reason": "the 1956 workshop named the field and starts the period of early optimism" } }, { "node1": "ai-p-john-mccarthy", "node2": "ai-a-dartmouth-workshop", "name1": "is organised by", "name2": "organises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-author of the proposal and the workshop's organiser who chose the name" } }, { "node1": "ai-p-marvin-minsky", "node2": "ai-a-dartmouth-workshop", "name1": "is organised by", "name2": "organises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-author of the 1955 proposal" } }, { "node1": "ai-p-claude-shannon", "node2": "ai-a-dartmouth-workshop", "name1": "is organised by", "name2": "organises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-author of the 1955 proposal" } }, { "node1": "ai-hist-p-allen-newell", "node2": "ai-hist-ms-logic-theorist", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "co-creator with Simon and Shaw" } }, { "node1": "ai-hist-p-herbert-simon", "node2": "ai-hist-ms-logic-theorist", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "co-creator with Newell and Shaw" } }, { "node1": "ai-hist-p-allen-newell", "node2": "ai-hist-o-carnegie-mellon", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "earned his PhD there under Simon and taught there" } }, { "node1": "ai-hist-p-herbert-simon", "node2": "ai-hist-o-carnegie-mellon", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "at Carnegie Mellon 1949 to 2001" } }, { "node1": "ai-hist-p-herbert-simon", "node2": "ai-hist-p-allen-newell", "name1": "is supervised by", "name2": "supervises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "mentorship", "Reason": "Newell earned his Carnegie Mellon PhD under Simon" } }, { "node1": "ai-p-frank-rosenblatt", "node2": "ai-a-perceptron", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "invented and led the perceptron project" } }, { "node1": "ai-o-cornell-aero-lab", "node2": "ai-a-perceptron", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "the laboratory where the perceptron was developed with Navy funding" } }, { "node1": "ai-p-frank-rosenblatt", "node2": "ai-o-cornell-aero-lab", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "led the cognitive systems section there" } }, { "node1": "ai-p-marvin-minsky", "node2": "ai-hist-pa-perceptrons-book", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author with Papert" } }, { "node1": "ai-a-perceptron", "node2": "ai-hist-pa-perceptrons-book", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the book analyses the limits of perceptrons and is dedicated to Rosenblatt" } }, { "node1": "ai-p-marvin-minsky", "node2": "ai-o-mit", "name1": "employs", "name2": "works at", "direction": 1, "properties": { "Relation": "composition", "Link kind": "employment", "Reason": "MIT faculty from 1958; co-founded the AI laboratory in 1959" } }, { "node1": "ai-a-transformer", "node2": "ai-era-7", "name1": "is opened by", "name2": "opens", "direction": 1, "properties": { "Relation": "causation", "Link kind": "era boundary", "Reason": "publication of the transformer paper starts the era" } }, { "node1": "ai-era-7", "node2": "ai-era-8", "name1": "follows", "name2": "precedes", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "era boundary", "Reason": "era 7 ends the day before ChatGPT starts era 8" } }, { "node1": "ai-o-google", "node2": "ai-a-transformer", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "all eight authors worked at Google when the paper appeared" } }, { "node1": "ai-p-ashish-vaswani", "node2": "ai-a-transformer", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first-listed author of the paper" } }, { "node1": "ai-scale-p-jakob-uszkoreit", "node2": "ai-a-transformer", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author who proposed the Transformer name" } }, { "node1": "ai-a-transformer", "node2": "ai-a-bert", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "BERT is an encoder-only transformer" } }, { "node1": "ai-a-transformer", "node2": "ai-scale-ms-gpt-1", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "GPT-1 is a 12-layer transformer decoder" } }, { "node1": "ai-o-google", "node2": "ai-a-bert", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "BERT came from Google AI" } }, { "node1": "ai-scale-p-jacob-devlin", "node2": "ai-a-bert", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "first author of the BERT paper" } }, { "node1": "ai-o-openai", "node2": "ai-scale-ms-gpt-1", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "OpenAI published GPT-1 on 11 June 2018" } }, { "node1": "ai-p-alec-radford", "node2": "ai-scale-ms-gpt-1", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "lead author of the GPT-1 paper" } }, { "node1": "ai-scale-ms-gpt-1", "node2": "ai-a-gpt-2", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "GPT-2 scales the GPT-1 design to 1.5 billion parameters" } }, { "node1": "ai-a-bert", "node2": "ai-scale-ms-gpt-1", "name1": "competes with", "name2": "competes with", "direction": 0, "properties": { "Relation": "comparison", "Link kind": "rivalry", "Reason": "BERT paper benchmarks against OpenAI GPT as the leading pretrained transformer" } }, { "node1": "ai-o-openai", "node2": "ai-a-gpt-2", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "OpenAI built and announced GPT-2" } }, { "node1": "ai-p-alec-radford", "node2": "ai-a-gpt-2", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "lead author of the GPT-2 paper" } }, { "node1": "ai-a-gpt-2", "node2": "ai-scale-re-gpt2-too-dangerous", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "critics answered OpenAI's decision to withhold the full model" } }, { "node1": "ai-scale-na-verge-gpt2-debate", "node2": "ai-scale-re-gpt2-too-dangerous", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers the researcher debate over withholding GPT-2" } }, { "node1": "ai-a-gpt-2", "node2": "ai-a-gpt-3", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "GPT-3 scales the GPT line to 175 billion parameters" } }, { "node1": "ai-scale-pa-scaling-laws", "node2": "ai-a-gpt-3", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "power-law results motivated training one very large model; same lab, four months later" } }, { "node1": "ai-scale-pa-scaling-laws", "node2": "ai-scale-pa-chinchilla", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "Chinchilla revisits the size-versus-data trade-off the 2020 study set out" } }, { "node1": "ai-o-openai", "node2": "ai-scale-pa-scaling-laws", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "authors include OpenAI researchers Brown, Radford, Amodei" } }, { "node1": "ai-p-dario-amodei", "node2": "ai-scale-pa-scaling-laws", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "listed co-author" } }, { "node1": "ai-p-alec-radford", "node2": "ai-scale-pa-scaling-laws", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "listed co-author" } }, { "node1": "ai-o-openai", "node2": "ai-a-gpt-3", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "OpenAI built and released GPT-3" } }, { "node1": "ai-a-gpt-3", "node2": "ai-a-v100", "name1": "is used to train", "name2": "is trained on", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "training compute", "Reason": "paper: all models were trained on V100 GPUs on a Microsoft-provided cluster" } }, { "node1": "ai-a-gpt-3", "node2": "ai-scale-ch-azure-openai-supercomputer", "name1": "is used to train", "name2": "is trained on", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "training compute", "Reason": "paper credits a Microsoft-provided cluster; the 19 May 2020 Azure machine was built for OpenAI" } }, { "node1": "ai-a-microsoft-openai-2019", "node2": "ai-scale-ch-azure-openai-supercomputer", "name1": "is made possible by", "name2": "makes possible", "direction": 1, "properties": { "Relation": "causation", "Link kind": "financing", "Reason": "partnership committed the two to build Azure AI supercomputing" } }, { "node1": "ai-a-microsoft-openai-2019", "node2": "ai-a-gpt-3", "name1": "is made possible by", "name2": "makes possible", "direction": 1, "properties": { "Relation": "causation", "Link kind": "financing", "Reason": "GPT-3 was trained on a Microsoft-provided cluster after the 2019 partnership" } }, { "node1": "ai-o-microsoft", "node2": "ai-a-microsoft-openai-2019", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Microsoft is the investor" } }, { "node1": "ai-a-microsoft-openai-2019", "node2": "ai-o-openai", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "OpenAI is the recipient" } }, { "node1": "ai-p-satya-nadella", "node2": "ai-o-microsoft", "name1": "is led by", "name2": "leads", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "leadership", "Reason": "chief executive of Microsoft" } }, { "node1": "ai-o-openai", "node2": "ai-scale-ev-openai-lp", "name1": "is organised by", "name2": "organises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "OpenAI created the capped-profit arm" } }, { "node1": "ai-scale-ev-openai-lp", "node2": "ai-a-microsoft-openai-2019", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Microsoft invested in OpenAI LP after the restructuring" } }, { "node1": "ai-a-gpt-3", "node2": "ai-scale-pr-openai-api", "name1": "is powered by", "name2": "powers", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "deployment", "Reason": "the API launched with GPT-3 as its model" } }, { "node1": "ai-o-openai", "node2": "ai-scale-pr-openai-api", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "OpenAI operates the API" } }, { "node1": "ai-a-gpt-3", "node2": "ai-scale-re-gpt3-bloviator", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "expert critique of GPT-3's comprehension" } }, { "node1": "ai-o-microsoft", "node2": "ai-scale-ch-azure-openai-supercomputer", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Microsoft built the machine in Azure" } }, { "node1": "ai-o-nvidia", "node2": "ai-a-v100", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Nvidia designed and announced the V100" } }, { "node1": "ai-a-gtx-580", "node2": "ai-a-v100", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "Nvidia's GPU line for deep learning moves from gaming Fermi to data-centre Volta" } }, { "node1": "ai-o-nvidia", "node2": "ai-a-gtx-580", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Nvidia launched the card on 9 November 2010" } }, { "node1": "ai-o-nvidia", "node2": "ai-a-cuda", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Nvidia created and ships CUDA" } }, { "node1": "ai-a-cuda", "node2": "ai-a-alexnet", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "AlexNet is written in CUDA and C++" } }, { "node1": "ai-p-jensen-huang", "node2": "ai-o-nvidia", "name1": "is led by", "name2": "leads", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "leadership", "Reason": "co-founder and chief executive" } }, { "node1": "ai-o-google", "node2": "ai-a-tpu", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Google designed the TPU and announced it at I/O 2016" } }, { "node1": "ai-scale-ms-alphazero", "node2": "ai-a-tpu", "name1": "is used to train", "name2": "is trained on", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "training compute", "Reason": "5,000 first-generation TPUs generated self-play games; 64 second-generation TPUs trained the networks" } }, { "node1": "ai-a-bert", "node2": "ai-a-tpu", "name1": "is used to train", "name2": "is trained on", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "training compute", "Reason": "BERT base trained on 4 Cloud TPUs (16 chips) for 4 days" } }, { "node1": "ai-a-v100", "node2": "ai-a-a100", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "Ampere is the successor of Volta" } }, { "node1": "ai-o-nvidia", "node2": "ai-a-a100", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Nvidia announced the A100 on 14 May 2020" } }, { "node1": "ai-a-a100", "node2": "ai-a-h100", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "Hopper follows Ampere in Nvidia's data-centre line" } }, { "node1": "ai-o-nvidia", "node2": "ai-a-h100", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Nvidia announced the H100 at GTC on 22 March 2022" } }, { "node1": "ai-a-h100", "node2": "ai-a-blackwell", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "Blackwell succeeds Hopper" } }, { "node1": "ai-o-nvidia", "node2": "ai-a-blackwell", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Nvidia announced Blackwell at GTC 2024" } }, { "node1": "ai-a-stable-diffusion", "node2": "ai-a-a100", "name1": "is used to train", "name2": "is trained on", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "training compute", "Reason": "trained on 256 A100 GPUs for about 150,000 GPU-hours" } }, { "node1": "ai-a-us-export-controls-2022", "node2": "ai-a-a100", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "the 7 October 2022 rule restricts sales of the A100-class chips to China" } }, { "node1": "ai-a-us-export-controls-2022", "node2": "ai-a-h100", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "the H100 was barred from sale to China in 2022" } }, { "node1": "ai-a-us-export-controls-2022", "node2": "ai-scale-ch-a800", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Nvidia built the A800 to stay under the new limit" } }, { "node1": "ai-o-nvidia", "node2": "ai-scale-ch-a800", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Nvidia produced the A800 for China" } }, { "node1": "ai-a-a100", "node2": "ai-scale-ch-a800", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "the A800 is the A100 with a slower interconnect for China" } }, { "node1": "ai-a-us-export-controls-2022", "node2": "ai-scale-ch-h800", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the H800 was built for China after the 2022 restrictions" } }, { "node1": "ai-a-h100", "node2": "ai-scale-ch-h800", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "the H800 is an H100 with reduced bandwidth for China" } }, { "node1": "ai-o-nvidia", "node2": "ai-scale-ch-h800", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Nvidia built the H800" } }, { "node1": "ai-scale-rg-export-update-2023", "node2": "ai-scale-ch-h800", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "the 17 October 2023 rule banned H800 exports to China" } }, { "node1": "ai-scale-rg-export-update-2023", "node2": "ai-scale-ch-h20", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the H20 was created after the October 2023 restrictions" } }, { "node1": "ai-o-nvidia", "node2": "ai-scale-ch-h20", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Nvidia built the H20 for China" } }, { "node1": "ai-scale-ch-ascend-910c", "node2": "ai-scale-ch-h20", "name1": "competes with", "name2": "competes with", "direction": 0, "properties": { "Relation": "comparison", "Link kind": "rivalry", "Reason": "Huawei's 910C filled the gap left by the H20 licence requirement" } }, { "node1": "ai-scale-o-huawei", "node2": "ai-scale-ch-ascend-910c", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Huawei designs the Ascend line" } }, { "node1": "ai-o-deepmind", "node2": "ai-scale-ms-alphago-zero", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "DeepMind built and published AlphaGo Zero" } }, { "node1": "ai-scale-ms-alphago-zero", "node2": "ai-scale-ms-alphazero", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "AlphaZero uses an approach similar to AlphaGo Zero" } }, { "node1": "ai-o-deepmind", "node2": "ai-scale-ms-alphazero", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "DeepMind posted and published AlphaZero" } }, { "node1": "ai-scale-ms-alphazero", "node2": "ai-scale-re-alphazero-chess-debate", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "chess masters and the Stockfish author reacted to the match conditions" } }, { "node1": "ai-o-deepmind", "node2": "ai-a-alphafold-2", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "DeepMind developed AlphaFold 2" } }, { "node1": "ai-p-john-jumper", "node2": "ai-a-alphafold-2", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "led the AlphaFold 2 project" } }, { "node1": "ai-a-alphafold-2", "node2": "ai-scale-ev-casp14", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "median GDT 92.4 and best prediction for 88 of 97 targets" } }, { "node1": "ai-a-alphafold-2", "node2": "ai-a-nobel-chemistry-2024", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "the prize recognised protein structure prediction, the work behind AlphaFold 2" } }, { "node1": "ai-scale-na-mit-alphafold", "node2": "ai-scale-ev-casp14", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "reports the CASP14 result" } }, { "node1": "ai-scale-na-mit-alphafold", "node2": "ai-a-alphafold-2", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers the AlphaFold system" } }, { "node1": "ai-o-deepmind", "node2": "ai-scale-pa-chinchilla", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "DeepMind authors Hoffmann, Borgeaud, Mensch and others" } }, { "node1": "ai-o-google", "node2": "ai-scale-pa-chain-of-thought", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Google researchers Wei, Zhou, Le and others" } }, { "node1": "ai-scale-pa-chain-of-thought", "node2": "ai-a-o1", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "reasoning models spend tokens on step-by-step reasoning, the idea chain-of-thought prompting introduced" } }, { "node1": "ai-a-gpt-3", "node2": "ai-a-instructgpt", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "InstructGPT fine-tunes GPT-3 with human feedback" } }, { "node1": "ai-o-openai", "node2": "ai-a-instructgpt", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "OpenAI authors Ouyang, Wu, Schulman, Leike, Lowe and others" } }, { "node1": "ai-a-instructgpt", "node2": "ai-a-chatgpt", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "the InstructGPT training workflow was used in the original version of ChatGPT" } }, { "node1": "ai-o-google", "node2": "ai-scale-ms-lamda", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Google unveiled LaMDA at I/O 2021" } }, { "node1": "ai-scale-ms-lamda", "node2": "ai-scale-re-lemoine-sentience", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Lemoine claimed LaMDA was sentient" } }, { "node1": "ai-scale-na-wapo-lemoine", "node2": "ai-scale-re-lemoine-sentience", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "first report of the claim" } }, { "node1": "ai-scale-pa-stochastic-parrots", "node2": "ai-scale-re-gebru-exit", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Google's objection to the draft led to Gebru's exit" } }, { "node1": "ai-scale-p-timnit-gebru", "node2": "ai-scale-pa-stochastic-parrots", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "co-author with Bender, McMillan-Major and Mitchell" } }, { "node1": "ai-scale-re-gebru-exit", "node2": "ai-o-google", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the dispute and the open letter were aimed at Google" } }, { "node1": "ai-scale-na-mit-gebru-paper", "node2": "ai-scale-pa-stochastic-parrots", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "summarises the paper's four risks" } }, { "node1": "ai-scale-na-mit-gebru-paper", "node2": "ai-scale-re-gebru-exit", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "reports on the dispute" } }, { "node1": "ai-scale-na-mit-gebru-paper", "node2": "ai-scale-p-timnit-gebru", "name1": "is quoted in", "name2": "quotes", "direction": 1, "properties": { "Relation": "information", "Link kind": "quotation", "Reason": "reports Gebru's account" } }, { "node1": "ai-o-openai", "node2": "ai-scale-ms-dall-e", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "OpenAI revealed DALL-E on 5 January 2021" } }, { "node1": "ai-a-gpt-3", "node2": "ai-scale-ms-dall-e", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "DALL-E is a version of GPT-3 modified to generate images" } }, { "node1": "ai-o-openai", "node2": "ai-scale-ms-clip", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "OpenAI released CLIP with DALL-E" } }, { "node1": "ai-p-alec-radford", "node2": "ai-scale-ms-clip", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "lead author of the CLIP paper" } }, { "node1": "ai-scale-ms-clip", "node2": "ai-scale-ms-dall-e", "name1": "is powered by", "name2": "powers", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "deployment", "Reason": "CLIP ranks DALL-E's outputs" } }, { "node1": "ai-scale-ms-dall-e", "node2": "ai-scale-ms-dall-e-2", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "DALL-E 2 replaces the autoregressive design with diffusion" } }, { "node1": "ai-o-openai", "node2": "ai-scale-ms-dall-e-2", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "OpenAI announced DALL-E 2 on 6 April 2022" } }, { "node1": "ai-scale-pa-ddpm", "node2": "ai-scale-ms-dall-e-2", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "DALL-E 2 uses a diffusion model" } }, { "node1": "ai-scale-pa-ddpm", "node2": "ai-scale-pa-latent-diffusion", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "latent diffusion moves the DDPM process into an autoencoder's latent space" } }, { "node1": "ai-scale-pa-latent-diffusion", "node2": "ai-a-stable-diffusion", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "Stable Diffusion originated from the Latent Diffusion research at LMU Munich" } }, { "node1": "ai-a-gpt-3", "node2": "ai-scale-ms-codex", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "Codex is a GPT model fine-tuned on GitHub code" } }, { "node1": "ai-o-openai", "node2": "ai-scale-ms-codex", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "OpenAI authors led by Mark Chen" } }, { "node1": "ai-scale-ms-codex", "node2": "ai-scale-pr-github-copilot", "name1": "is powered by", "name2": "powers", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "deployment", "Reason": "Copilot was initially powered by OpenAI Codex" } }, { "node1": "ai-o-microsoft", "node2": "ai-scale-pr-github-copilot", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "GitHub, a Microsoft subsidiary, launched Copilot with OpenAI" } }, { "node1": "ai-scale-pr-github-copilot", "node2": "ai-scale-re-copilot-class-action", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "developers sued over Copilot's use of open-source code" } }, { "node1": "ai-scale-re-copilot-class-action", "node2": "ai-o-microsoft", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Microsoft is a defendant" } }, { "node1": "ai-scale-re-copilot-class-action", "node2": "ai-o-openai", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "OpenAI is a defendant" } }, { "node1": "ai-scale-p-david-holz", "node2": "ai-scale-pr-midjourney", "name1": "is created by", "name2": "creates", "direction": 1, "properties": { "Relation": "causation", "Link kind": "creation", "Reason": "founder of Midjourney" } }, { "node1": "ai-scale-pa-ddpm", "node2": "ai-scale-pr-midjourney", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "diffusion models underlie modern image generators" } }, { "node1": "ai-scale-pr-midjourney", "node2": "ai-scale-ev-colorado-fair", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the winning entry was made with Midjourney" } }, { "node1": "ai-scale-ev-colorado-fair", "node2": "ai-scale-re-artists-colorado", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "artists objected after the win became public" } }, { "node1": "ai-scale-na-nyt-colorado", "node2": "ai-scale-re-artists-colorado", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers the artists' objections" } }, { "node1": "ai-scale-na-nyt-colorado", "node2": "ai-scale-ev-colorado-fair", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers the prize" } }, { "node1": "ai-o-stability-ai", "node2": "ai-a-stable-diffusion", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Stability AI supplied compute and released the model" } }, { "node1": "ai-p-emad-mostaque", "node2": "ai-o-stability-ai", "name1": "is founded by", "name2": "founds", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founder with Cyrus Hodes" } }, { "node1": "ai-scale-fr-stability-101m", "node2": "ai-o-stability-ai", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "round raised by Stability AI" } }, { "node1": "ai-a-stable-diffusion", "node2": "ai-scale-fr-stability-101m", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the round came two months after the model's release" } }, { "node1": "ai-p-garry-kasparov", "node2": "ai-scale-re-alphazero-chess-debate", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "Kasparov called AlphaZero a remarkable achievement" } }, { "node1": "ai-scale-p-timnit-gebru", "node2": "ai-scale-re-gebru-exit", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "her account of being forced out started the dispute" } }, { "node1": "ai-o-china-state-council", "node2": "ai-a-china-ai-plan-2017", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "the State Council issued the plan on 20 July 2017" } }, { "node1": "ai-a-alphago-lee-sedol", "node2": "ai-a-china-ai-plan-2017", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "AlphaGo's victories raised China's focus on AI, as an academic analysis and Kai-Fu Lee's Sputnik remark describe" } }, { "node1": "ai-o-european-commission", "node2": "ai-a-eu-ai-act-proposal", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "the Commission tabled the proposal on 21 April 2021" } }, { "node1": "ai-a-eu-ai-act-proposal", "node2": "ai-scale-re-eu-proposal-reactions", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "rights groups and industry reacted on the day of publication" } }, { "node1": "ai-scale-na-euronews-ai-act", "node2": "ai-a-eu-ai-act-proposal", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers the draft rules" } }, { "node1": "ai-scale-na-euronews-ai-act", "node2": "ai-scale-re-eu-proposal-reactions", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "records the critics' quotes" } }, { "node1": "ai-o-us-commerce-bis", "node2": "ai-a-us-export-controls-2022", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "BIS announced the rules on 7 October 2022" } }, { "node1": "ai-a-us-export-controls-2022", "node2": "ai-scale-rg-export-update-2023", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the October 2023 rules modified and reinforced the restrictions first released on 7 October 2022" } }, { "node1": "ai-o-us-commerce-bis", "node2": "ai-scale-rg-export-update-2023", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "BIS released the three rules on 17 October 2023" } }, { "node1": "ai-p-joe-biden", "node2": "ai-scale-rg-chips-act", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "signed the law on 9 August 2022" } }, { "node1": "ai-scale-rg-chips-act", "node2": "ai-a-us-export-controls-2022", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "domestic chip subsidies and export restrictions were two halves of the same China-facing strategy" } }, { "node1": "ai-o-us-commerce-bis", "node2": "ai-scale-rg-diffusion-rule-2025", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "BIS issued the rule in January 2025" } }, { "node1": "ai-o-us-commerce-bis", "node2": "ai-scale-rg-diffusion-rescission-2025", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "BIS rescinded the rule on 13 May 2025" } }, { "node1": "ai-scale-rg-diffusion-rescission-2025", "node2": "ai-scale-rg-diffusion-rule-2025", "name1": "is revoked by", "name2": "revokes", "direction": 1, "properties": { "Relation": "causation", "Link kind": "revocation", "Reason": "the rescission replaced the Biden-era rule before it took effect" } }, { "node1": "ai-scale-rg-diffusion-rescission-2025", "node2": "ai-scale-ch-ascend-910c", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "guidance said using Huawei Ascend chips anywhere violates US export controls" } }, { "node1": "ai-scale-na-hill-rescission", "node2": "ai-scale-rg-diffusion-rescission-2025", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers the rescission" } }, { "node1": "ai-scale-na-hill-rescission", "node2": "ai-scale-rg-diffusion-rule-2025", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "describes the rule's cap on chip sales" } }, { "node1": "ai-o-us-commerce-bis", "node2": "ai-scale-rg-h20-licence-2025", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "the licence requirement took effect on 9 April 2025" } }, { "node1": "ai-scale-rg-h20-licence-2025", "node2": "ai-scale-ch-h20", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "licence needed to sell the H20 to China" } }, { "node1": "ai-scale-rg-h20-licence-2025", "node2": "ai-scale-ch-ascend-910c", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Chinese cloud firms turned to Huawei's 910C as H20 substitute" } }, { "node1": "ai-scale-rg-h20-reversal-2025", "node2": "ai-scale-rg-h20-licence-2025", "name1": "is revoked by", "name2": "revokes", "direction": 1, "properties": { "Relation": "causation", "Link kind": "revocation", "Reason": "the US reversed the April H20 ban" } }, { "node1": "ai-scale-na-decoder-h20", "node2": "ai-scale-rg-h20-reversal-2025", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers the reversal" } }, { "node1": "ai-o-us-commerce-bis", "node2": "ai-scale-rg-h200-china-2026", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "BIS issued the final rule on 13 January 2026" } }, { "node1": "ai-scale-rg-h200-china-2026", "node2": "ai-o-nvidia", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "covers sales of H200-class chips" } }, { "node1": "ai-scale-rg-h200-china-2026", "node2": "ai-scale-o-amd", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "covers MI325X-class chips" } }, { "node1": "ai-scale-na-fnnews-h200", "node2": "ai-scale-rg-h200-china-2026", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "reports on how little has shipped under the licences" } }, { "node1": "ai-scale-mm-nvidia-guidance-2023", "node2": "ai-o-nvidia", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Nvidia shares rose about 24% on 25 May 2023" } }, { "node1": "ai-a-chatgpt", "node2": "ai-scale-mm-nvidia-guidance-2023", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "the report revealed the surge in chip demand set off by ChatGPT's launch" } }, { "node1": "ai-a-h100", "node2": "ai-scale-mm-nvidia-guidance-2023", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "H100 demand drove the data-centre forecast" } }, { "node1": "ai-scale-na-nvidia-q1-release", "node2": "ai-scale-mm-nvidia-guidance-2023", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "the release carried the forecast that moved the stock" } }, { "node1": "ai-scale-na-nvidia-q1-release", "node2": "ai-p-jensen-huang", "name1": "is quoted in", "name2": "quotes", "direction": 1, "properties": { "Relation": "information", "Link kind": "quotation", "Reason": "quotes Huang on accelerated computing and generative AI" } }, { "node1": "ai-scale-mm-nvidia-1t-2023", "node2": "ai-o-nvidia", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Nvidia touched a $1 trillion value on 30 May 2023" } }, { "node1": "ai-a-h100", "node2": "ai-scale-mm-nvidia-1t-2023", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "AI demand for Nvidia's data-centre chips pushed the stock toward $1 trillion" } }, { "node1": "ai-scale-na-yahoo-nvidia-1t", "node2": "ai-scale-mm-nvidia-1t-2023", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers the $1 trillion touch" } }, { "node1": "ai-scale-mm-nvidia-record-gain-2024", "node2": "ai-o-nvidia", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Nvidia shares rose 16.4% on 22 February 2024" } }, { "node1": "ai-a-h100", "node2": "ai-scale-mm-nvidia-record-gain-2024", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "results showed strong demand for AI computing chips led by the H100" } }, { "node1": "ai-scale-na-business-reporter-277", "node2": "ai-scale-mm-nvidia-record-gain-2024", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "Reuters report on the record gain" } }, { "node1": "ai-scale-mm-nvidia-most-valuable-2024", "node2": "ai-o-nvidia", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Nvidia shares rose 3.5% on 18 June 2024" } }, { "node1": "ai-a-h100", "node2": "ai-scale-mm-nvidia-most-valuable-2024", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "voracious demand for Nvidia's chips from Microsoft, Meta and Google" } }, { "node1": "ai-scale-na-aljazeera-nvidia-3t", "node2": "ai-scale-mm-nvidia-most-valuable-2024", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers Nvidia passing Microsoft" } }, { "node1": "ai-scale-mm-nvidia-4t-2025", "node2": "ai-o-nvidia", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Nvidia shares rose to a $4 trillion value on 9 July 2025" } }, { "node1": "ai-a-blackwell", "node2": "ai-scale-mm-nvidia-4t-2025", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "demand for Nvidia's latest data-centre chips, which were sold out for 2025" } }, { "node1": "ai-scale-mm-nvidia-5t-2025", "node2": "ai-o-nvidia", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Nvidia shares rose to a $5 trillion value on 29 October 2025" } }, { "node1": "ai-a-blackwell", "node2": "ai-scale-mm-nvidia-5t-2025", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Huang's $500 billion order forecast covers Nvidia's AI chips" } }, { "node1": "ai-scale-na-entrepreneur-nvidia-5t", "node2": "ai-scale-mm-nvidia-5t-2025", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "covers the $5 trillion milestone" } }, { "node1": "ai-scale-mm-amd-openai-2025", "node2": "ai-scale-o-amd", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "AMD shares rose after the OpenAI agreement on 6 October 2025" } }, { "node1": "ai-a-nvidia-deepseek-selloff", "node2": "ai-o-nvidia", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Nvidia lost $593 billion of market value on 27 January 2025" } }, { "node1": "ai-p-dario-amodei", "node2": "ai-o-anthropic", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "co-founded Anthropic in January 2021" } }, { "node1": "ai-o-tsmc", "node2": "ai-a-h100", "name1": "is fabricated by", "name2": "fabricates", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "supply", "Reason": "the H100 is built on TSMC's 4N process" } }, { "node1": "ai-o-tsmc", "node2": "ai-a-blackwell", "name1": "is fabricated by", "name2": "fabricates", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "supply", "Reason": "a Blackwell design flaw was fixed with TSMC" } }, { "node1": "ai-scale-pa-human-preferences", "node2": "ai-a-instructgpt", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "InstructGPT applies learning from human preference rankings to language models" } }, { "node1": "ai-p-dario-amodei", "node2": "ai-scale-pa-human-preferences", "name1": "is written by", "name2": "writes", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "listed co-author" } }, { "node1": "ai-scale-ev-ke-jie", "node2": "ai-a-china-ai-plan-2017", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Kai-Fu Lee called the match China's Sputnik moment; the plan followed in July" } }, { "node1": "ai-a-tpu", "node2": "ai-scale-ev-ke-jie", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "AlphaGo Master ran on four TPUs" } }, { "node1": "ai-o-deepmind", "node2": "ai-scale-ev-ke-jie", "name1": "is organised by", "name2": "organises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "DeepMind and Google ran the summit with Chinese partners" } }, { "node1": "ai-a-bert", "node2": "ai-scale-pr-google-search-bert", "name1": "is powered by", "name2": "powers", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "deployment", "Reason": "Google used BERT to interpret search queries from 25 October 2019" } }, { "node1": "ai-o-google", "node2": "ai-scale-pr-google-search-bert", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Google operates Search" } }, { "node1": "ai-o-nvidia", "node2": "ai-scale-fr-nvidia-arm", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Nvidia was the buyer" } }, { "node1": "ai-o-softbank", "node2": "ai-scale-fr-nvidia-arm", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "SoftBank was the seller and would have taken a stake in Nvidia" } }, { "node1": "ai-o-openai", "node2": "ai-a-chatgpt", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "OpenAI built and launched ChatGPT" } }, { "node1": "ai-o-openai", "node2": "ai-boom-ms-gpt-35", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "OpenAI trained the GPT-3.5 series" } }, { "node1": "ai-boom-ms-gpt-35", "node2": "ai-a-chatgpt", "name1": "is powered by", "name2": "powers", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "deployment", "Reason": "GPT-3.5 was the first model used in ChatGPT" } }, { "node1": "ai-a-instructgpt", "node2": "ai-boom-ms-gpt-35", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "GPT-3.5 and ChatGPT used the human-feedback fine-tuning method set out in InstructGPT" } }, { "node1": "ai-a-chatgpt", "node2": "ai-boom-re-google-code-red", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Google treated ChatGPT as a threat to search and declared a code red" } }, { "node1": "ai-o-google", "node2": "ai-boom-re-google-code-red", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "response", "Reason": "Google management issued the alert" } }, { "node1": "ai-boom-na-nyt-code-red", "node2": "ai-boom-re-google-code-red", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "the article broke the code red story" } }, { "node1": "ai-p-sundar-pichai", "node2": "ai-o-google", "name1": "is led by", "name2": "leads", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "leadership", "Reason": "CEO of Google and Alphabet" } }, { "node1": "ai-boom-re-google-code-red", "node2": "ai-boom-pr-bard", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the code red led Google to rush Bard to announcement" } }, { "node1": "ai-o-google", "node2": "ai-boom-pr-bard", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Google built Bard on LaMDA" } }, { "node1": "ai-boom-pr-bard", "node2": "ai-boom-mm-alphabet-bard-demo", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "the Bard demo error was followed by the share fall" } }, { "node1": "ai-boom-mm-alphabet-bard-demo", "node2": "ai-o-google", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Alphabet (GOOGL) share price fell" } }, { "node1": "ai-a-chatgpt", "node2": "ai-boom-fr-microsoft-openai-2023", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "financing", "Reason": "Microsoft extended its OpenAI investment weeks after ChatGPT's launch" } }, { "node1": "ai-o-microsoft", "node2": "ai-boom-fr-microsoft-openai-2023", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Microsoft is the investor" } }, { "node1": "ai-boom-fr-microsoft-openai-2023", "node2": "ai-o-openai", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "OpenAI is the recipient" } }, { "node1": "ai-boom-fr-microsoft-openai-2023", "node2": "ai-boom-pr-bing-chat", "name1": "is made possible by", "name2": "makes possible", "direction": 1, "properties": { "Relation": "causation", "Link kind": "financing", "Reason": "the partnership put OpenAI models into Microsoft products, first Bing Chat" } }, { "node1": "ai-o-microsoft", "node2": "ai-boom-pr-bing-chat", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Microsoft built Bing Chat" } }, { "node1": "ai-boom-pr-bing-chat", "node2": "ai-boom-pr-bard", "name1": "competes with", "name2": "competes with", "direction": 0, "properties": { "Relation": "comparison", "Link kind": "rivalry", "Reason": "Google rushed Bard to pre-empt Microsoft's Bing event" } }, { "node1": "ai-a-chatgpt", "node2": "ai-era-8", "name1": "is opened by", "name2": "opens", "direction": 1, "properties": { "Relation": "causation", "Link kind": "era boundary", "Reason": "ChatGPT's launch starts the generative AI boom" } }, { "node1": "ai-a-o1", "node2": "ai-era-9", "name1": "is opened by", "name2": "opens", "direction": 1, "properties": { "Relation": "causation", "Link kind": "era boundary", "Reason": "o1 is the first widely covered reasoning model and starts the period by convention" } }, { "node1": "ai-era-8", "node2": "ai-era-9", "name1": "follows", "name2": "precedes", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "era boundary", "Reason": "the boom era is followed by the reasoning-model era" } }, { "node1": "ai-o-meta", "node2": "ai-a-llama", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Meta released LLaMA" } }, { "node1": "ai-p-mark-zuckerberg", "node2": "ai-o-meta", "name1": "is led by", "name2": "leads", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "leadership", "Reason": "co-founder, chairman and CEO of Meta" } }, { "node1": "ai-o-openai", "node2": "ai-a-gpt-4", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "OpenAI built GPT-4" } }, { "node1": "ai-boom-ms-gpt-35", "node2": "ai-a-gpt-4", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "GPT-4 followed GPT-3.5 in OpenAI's line" } }, { "node1": "ai-a-gpt-4", "node2": "ai-a-o1", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "o1 continued OpenAI's GPT-4-class line with added reasoning training" } }, { "node1": "ai-a-gpt-4", "node2": "ai-boom-pa-sparks-agi", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Microsoft researchers tested an early GPT-4 and wrote the Sparks paper" } }, { "node1": "ai-o-microsoft", "node2": "ai-boom-pa-sparks-agi", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "authorship", "Reason": "Microsoft Research authors" } }, { "node1": "ai-a-gpt-4", "node2": "ai-a-pause-letter", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the letter asked for a pause on systems more powerful than GPT-4 and appeared a week after its release" } }, { "node1": "ai-o-future-of-life", "node2": "ai-a-pause-letter", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "FLI published the letter" } }, { "node1": "ai-p-elon-musk", "node2": "ai-a-pause-letter", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "Musk was a prominent signatory" } }, { "node1": "ai-p-yoshua-bengio", "node2": "ai-a-pause-letter", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "Bengio signed the letter" } }, { "node1": "ai-boom-na-guardian-pause", "node2": "ai-a-pause-letter", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "Guardian reported Musk joining the call" } }, { "node1": "ai-boom-na-guardian-pause", "node2": "ai-p-elon-musk", "name1": "is quoted in", "name2": "quotes", "direction": 1, "visibility": false, "properties": { "Relation": "information", "Link kind": "quotation", "Reason": "article centres on Musk signing" } }, { "node1": "ai-a-pause-letter", "node2": "ai-a-eo-14110", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the letter is credited with raising government urgency before the US executive order" } }, { "node1": "ai-a-chatgpt", "node2": "ai-boom-rg-italy-chatgpt-ban", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the Garante acted against ChatGPT over data protection" } }, { "node1": "ai-o-italy-garante", "node2": "ai-boom-rg-italy-chatgpt-ban", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "the Garante issued the order" } }, { "node1": "ai-boom-na-bloomberg-italy-restore", "node2": "ai-boom-rg-italy-chatgpt-ban", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "Bloomberg reported the restoration" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-boom-re-hinton-leaves-google", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "Hinton announced his resignation and his warning" } }, { "node1": "ai-a-gpt-4", "node2": "ai-boom-re-hinton-leaves-google", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Hinton cited the pace of chatbot progress after GPT-4" } }, { "node1": "ai-boom-na-nyt-hinton", "node2": "ai-boom-re-hinton-leaves-google", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "NYT broke the resignation" } }, { "node1": "ai-boom-na-nyt-hinton", "node2": "ai-p-geoffrey-hinton", "name1": "is quoted in", "name2": "quotes", "direction": 1, "properties": { "Relation": "information", "Link kind": "quotation", "Reason": "interview with Hinton" } }, { "node1": "ai-boom-re-hinton-leaves-google", "node2": "ai-boom-re-cais-statement", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "Hinton, newly free to speak, signed the CAIS statement a month later" } }, { "node1": "ai-o-center-for-ai-safety", "node2": "ai-boom-re-cais-statement", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "CAIS published the statement" } }, { "node1": "ai-p-geoffrey-hinton", "node2": "ai-boom-re-cais-statement", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "signatory" } }, { "node1": "ai-p-sam-altman", "node2": "ai-boom-re-cais-statement", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "signatory" } }, { "node1": "ai-p-dario-amodei", "node2": "ai-boom-re-cais-statement", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "signatory" } }, { "node1": "ai-boom-na-verge-extinction", "node2": "ai-boom-re-cais-statement", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "The Verge reported the statement" } }, { "node1": "ai-boom-re-cais-statement", "node2": "ai-a-bletchley", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the statement fed the UK government's decision to host a frontier-AI safety summit" } }, { "node1": "ai-a-chatgpt", "node2": "ai-boom-re-wga-strike", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "generative AI use by studios was one of the guild's demands" } }, { "node1": "ai-o-wga", "node2": "ai-boom-re-wga-strike", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "the union called the strike" } }, { "node1": "ai-boom-re-wga-strike", "node2": "ai-boom-re-wga-deal", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the strike ended with a contract containing AI terms" } }, { "node1": "ai-boom-na-thr-strike", "node2": "ai-boom-re-wga-strike", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "THR reported the strike call" } }, { "node1": "ai-boom-na-variety-wga-deal", "node2": "ai-boom-re-wga-deal", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "Variety reported the deal" } }, { "node1": "ai-a-chatgpt", "node2": "ai-boom-mm-chegg", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Chegg said ChatGPT was hurting its customer growth and shares fell" } }, { "node1": "ai-a-chatgpt", "node2": "ai-boom-rg-china-genai-measures", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "China's rules govern services like ChatGPT" } }, { "node1": "ai-o-cac", "node2": "ai-boom-rg-china-genai-measures", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "CAC led the issuing of the measures" } }, { "node1": "ai-a-altman-ouster", "node2": "ai-o-openai", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the board crisis concerned OpenAI's leadership and governance" } }, { "node1": "ai-o-white-house", "node2": "ai-a-eo-14110", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "issued by the Biden administration" } }, { "node1": "ai-p-joe-biden", "node2": "ai-a-eo-14110", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "signed on 30 October 2023" } }, { "node1": "ai-a-eo-14110-revoked", "node2": "ai-a-eo-14110", "name1": "is revoked by", "name2": "revokes", "direction": 1, "properties": { "Relation": "causation", "Link kind": "revocation", "Reason": "rescinded the Biden order on 20 January 2025" } }, { "node1": "ai-p-rishi-sunak", "node2": "ai-a-bletchley", "name1": "is organised by", "name2": "organises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "Sunak hosted the summit" } }, { "node1": "ai-a-bletchley", "node2": "ai-boom-rg-bletchley-declaration", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "regulation", "Reason": "the declaration was agreed at the summit" } }, { "node1": "ai-p-rishi-sunak", "node2": "ai-boom-rg-bletchley-declaration", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "UK was the host signatory" } }, { "node1": "ai-boom-na-guardian-bletchley", "node2": "ai-boom-rg-bletchley-declaration", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "Guardian reported the signing" } }, { "node1": "ai-o-deepmind", "node2": "ai-boom-ms-gemini-1", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Google DeepMind built Gemini" } }, { "node1": "ai-boom-pr-bard", "node2": "ai-boom-ms-gemini-1", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "Bard was later renamed Gemini" } }, { "node1": "ai-boom-ms-gemini-1", "node2": "ai-a-gpt-4", "name1": "competes with", "name2": "competes with", "direction": 0, "properties": { "Relation": "comparison", "Link kind": "rivalry", "Reason": "Google said Gemini Ultra beat GPT-4 on benchmarks" } }, { "node1": "ai-o-amazon", "node2": "ai-boom-fr-amazon-anthropic-2023", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Amazon is the investor" } }, { "node1": "ai-boom-fr-amazon-anthropic-2023", "node2": "ai-o-anthropic", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "Anthropic is the recipient" } }, { "node1": "ai-o-amazon", "node2": "ai-boom-fr-amazon-anthropic-2024", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Amazon is the investor" } }, { "node1": "ai-boom-fr-amazon-anthropic-2024", "node2": "ai-o-anthropic", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "Anthropic is the recipient" } }, { "node1": "ai-a-chatgpt", "node2": "ai-a-nyt-v-openai", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "the suit alleges ChatGPT outputs and training copied Times articles" } }, { "node1": "ai-o-new-york-times", "node2": "ai-a-nyt-v-openai", "name1": "is filed by", "name2": "files", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "lawsuit", "Reason": "the Times is the plaintiff" } }, { "node1": "ai-a-nyt-v-openai", "node2": "ai-o-openai", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "OpenAI is the defendant" } }, { "node1": "ai-a-nyt-v-openai", "node2": "ai-o-microsoft", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "Microsoft is a co-defendant" } }, { "node1": "ai-o-openai", "node2": "ai-boom-ms-sora", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "OpenAI built Sora" } }, { "node1": "ai-a-eu-ai-act-proposal", "node2": "ai-a-eu-ai-act", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "the 2021 Commission proposal became the 2024 Act" } }, { "node1": "ai-a-chatgpt", "node2": "ai-a-eu-ai-act", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "rules for general-purpose AI models were added to the Act after ChatGPT" } }, { "node1": "ai-boom-na-cnbc-eu-vote", "node2": "ai-a-eu-ai-act", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "CNBC reported the Parliament vote" } }, { "node1": "ai-p-elon-musk", "node2": "ai-boom-re-musk-sues-openai", "name1": "is filed by", "name2": "files", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "lawsuit", "Reason": "Musk is the plaintiff" } }, { "node1": "ai-boom-re-musk-sues-openai", "node2": "ai-o-openai", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "OpenAI is the defendant" } }, { "node1": "ai-o-microsoft", "node2": "ai-boom-fr-openai-6-6bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "participated in the round" } }, { "node1": "ai-o-nvidia", "node2": "ai-boom-fr-openai-6-6bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "participated in the round" } }, { "node1": "ai-o-softbank", "node2": "ai-boom-fr-openai-6-6bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "participated in the round" } }, { "node1": "ai-boom-fr-openai-6-6bn", "node2": "ai-o-openai", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "OpenAI is the recipient" } }, { "node1": "ai-o-openai", "node2": "ai-a-o1", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "OpenAI built o1" } }, { "node1": "ai-p-demis-hassabis", "node2": "ai-a-nobel-chemistry-2024", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "laureate" } }, { "node1": "ai-p-john-jumper", "node2": "ai-a-nobel-chemistry-2024", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "laureate" } }, { "node1": "ai-o-deepseek", "node2": "ai-boom-ms-deepseek-v3", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "DeepSeek released V3" } }, { "node1": "ai-boom-ms-deepseek-v3", "node2": "ai-a-deepseek-r1", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "R1 was trained from the V3 base" } }, { "node1": "ai-o-deepseek", "node2": "ai-a-deepseek-r1", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "DeepSeek built and released R1" } }, { "node1": "ai-a-o1", "node2": "ai-a-deepseek-r1", "name1": "competes with", "name2": "competes with", "direction": 0, "properties": { "Relation": "comparison", "Link kind": "rivalry", "Reason": "R1 was released as an open-weight rival to OpenAI's reasoning model" } }, { "node1": "ai-a-deepseek-r1", "node2": "ai-a-nvidia-deepseek-selloff", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Nvidia fell about 17 percent on 27 January after R1 and the DeepSeek app spread" } }, { "node1": "ai-boom-na-nyt-deepseek", "node2": "ai-a-deepseek-r1", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "NYT explainer on DeepSeek" } }, { "node1": "ai-boom-na-cnbc-deepseek", "node2": "ai-a-deepseek-r1", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "CNBC reported the app topping the App Store" } }, { "node1": "ai-boom-na-bbc-deepseek-shares", "node2": "ai-a-nvidia-deepseek-selloff", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "BBC reported the day after the sell-off" } }, { "node1": "ai-a-deepseek-r1", "node2": "ai-boom-rg-italy-deepseek-block", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the Garante acted three days after the app went viral" } }, { "node1": "ai-o-italy-garante", "node2": "ai-boom-rg-italy-deepseek-block", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "the Garante issued the order" } }, { "node1": "ai-boom-rg-italy-deepseek-block", "node2": "ai-o-deepseek", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "the order restricts processing by DeepSeek's operators" } }, { "node1": "ai-boom-rg-italy-chatgpt-ban", "node2": "ai-boom-rg-italy-deepseek-block", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "the Garante repeated its 2023 approach against another chatbot" } }, { "node1": "ai-a-deepseek-r1", "node2": "ai-boom-re-trump-deepseek-wakeup", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the US president reacted to the DeepSeek shock" } }, { "node1": "ai-p-donald-trump", "node2": "ai-boom-re-trump-deepseek-wakeup", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "Trump made the remark" } }, { "node1": "ai-o-softbank", "node2": "ai-a-stargate", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "SoftBank is a lead partner and chairman Son leads the venture" } }, { "node1": "ai-o-openai", "node2": "ai-a-stargate", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "OpenAI is an equity partner and the main customer" } }, { "node1": "ai-boom-o-oracle", "node2": "ai-a-stargate", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Oracle is an equity partner and technology partner" } }, { "node1": "ai-boom-na-reuters-stargate", "node2": "ai-a-stargate", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "Reuters reported the announcement" } }, { "node1": "ai-p-masayoshi-son", "node2": "ai-o-softbank", "name1": "is led by", "name2": "leads", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "leadership", "Reason": "founder, chairman and CEO of SoftBank" } }, { "node1": "ai-p-donald-trump", "node2": "ai-a-eo-14110-revoked", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "signed on his first day in office" } }, { "node1": "ai-o-white-house", "node2": "ai-a-eo-14110-revoked", "name1": "is enacted by", "name2": "enacts", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "issued by the Trump administration" } }, { "node1": "ai-a-eo-14110-revoked", "node2": "ai-boom-rg-eo-14179", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "regulation", "Reason": "EO 14179 followed three days later to set the replacement policy" } }, { "node1": "ai-p-donald-trump", "node2": "ai-boom-rg-eo-14179", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "signed on 23 January 2025" } }, { "node1": "ai-boom-rg-eo-14179", "node2": "ai-boom-rg-ai-action-plan", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "regulation", "Reason": "EO 14179 directed the development of the action plan" } }, { "node1": "ai-o-white-house", "node2": "ai-boom-rg-ai-action-plan", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "published by the White House" } }, { "node1": "ai-a-bletchley", "node2": "ai-boom-ev-paris-summit", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "the Paris summit continued the series after Bletchley and Seoul" } }, { "node1": "ai-o-anthropic", "node2": "ai-boom-pr-claude-code", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Anthropic built Claude Code" } }, { "node1": "ai-o-softbank", "node2": "ai-boom-fr-softbank-openai-40bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "SoftBank led the round" } }, { "node1": "ai-boom-fr-softbank-openai-40bn", "node2": "ai-o-openai", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "OpenAI is the recipient" } }, { "node1": "ai-o-meta", "node2": "ai-boom-fr-meta-scale-ai", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Meta bought a 49 percent non-voting stake" } }, { "node1": "ai-o-openai", "node2": "ai-boom-ms-gpt-5", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "OpenAI built GPT-5" } }, { "node1": "ai-a-gpt-4", "node2": "ai-boom-ms-gpt-5", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "GPT-5 is the successor to the GPT-4 line" } }, { "node1": "ai-boom-ms-gpt-5", "node2": "ai-a-chatgpt", "name1": "is powered by", "name2": "powers", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "deployment", "Reason": "GPT-5 became the default model in ChatGPT" } }, { "node1": "ai-a-eu-ai-act", "node2": "ai-boom-rg-ai-act-prohibitions", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "first phase of the Act to apply" } }, { "node1": "ai-boom-rg-ai-act-prohibitions", "node2": "ai-boom-rg-ai-act-gpai", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "GPAI duties applied six months after prohibitions" } }, { "node1": "ai-boom-rg-ai-act-gpai", "node2": "ai-boom-rg-eu-digital-omnibus", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "regulation", "Reason": "the omnibus amended the Act's schedule and AI Office powers after the GPAI stage" } }, { "node1": "ai-boom-na-lemonde-aiact", "node2": "ai-boom-rg-ai-act-gpai", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "Le Monde reported creators' protest on the day the duties applied" } }, { "node1": "ai-boom-re-bartz-settlement", "node2": "ai-o-anthropic", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "Anthropic is the defendant" } }, { "node1": "ai-boom-na-npr-bartz", "node2": "ai-boom-re-bartz-settlement", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "NPR reported the settlement" } }, { "node1": "ai-boom-fr-anthropic-13bn", "node2": "ai-o-anthropic", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "Anthropic is the recipient" } }, { "node1": "ai-boom-fr-amazon-anthropic-2024", "node2": "ai-boom-fr-anthropic-13bn", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "the next large Anthropic round" } }, { "node1": "ai-boom-fr-anthropic-13bn", "node2": "ai-boom-fr-anthropic-30bn", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "the next Anthropic round (Series G)" } }, { "node1": "ai-boom-fr-oracle-openai-300bn", "node2": "ai-boom-mm-oracle-openai", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Oracle shares surged after the Journal reported the contract" } }, { "node1": "ai-boom-mm-oracle-openai", "node2": "ai-boom-o-oracle", "name1": "has shares moved by", "name2": "moves shares of", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "Oracle (ORCL) share price rose" } }, { "node1": "ai-a-stargate", "node2": "ai-boom-fr-oracle-openai-300bn", "name1": "is made possible by", "name2": "makes possible", "direction": 1, "properties": { "Relation": "causation", "Link kind": "financing", "Reason": "the Oracle compute contract is part of the Stargate build-out" } }, { "node1": "ai-o-nvidia", "node2": "ai-boom-fr-nvidia-openai-100bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Nvidia is the investor" } }, { "node1": "ai-boom-fr-nvidia-openai-100bn", "node2": "ai-o-openai", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "OpenAI is the recipient" } }, { "node1": "ai-boom-rg-ca-sb53", "node2": "ai-o-openai", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "OpenAI is a large frontier developer within the bill's thresholds" } }, { "node1": "ai-a-eo-14110-revoked", "node2": "ai-boom-rg-ca-sb53", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "regulation", "Reason": "California legislated frontier-AI safety rules after federal rules were rolled back" } }, { "node1": "ai-boom-ms-gemini-1", "node2": "ai-boom-ms-gemini-3", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "third generation of the Gemini line" } }, { "node1": "ai-o-deepmind", "node2": "ai-boom-ms-gemini-3", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Google DeepMind built Gemini 3" } }, { "node1": "ai-boom-ms-gemini-3", "node2": "ai-boom-re-openai-code-red", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "OpenAI declared a code red weeks after Gemini 3 (accounts differ on directness)" } }, { "node1": "ai-p-sam-altman", "node2": "ai-boom-re-openai-code-red", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "response", "Reason": "Altman's internal memo" } }, { "node1": "ai-boom-fr-disney-openai", "node2": "ai-o-openai", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "OpenAI is the recipient" } }, { "node1": "ai-boom-ms-sora", "node2": "ai-boom-fr-disney-openai", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "financing", "Reason": "the Disney licence was for characters on Sora 2" } }, { "node1": "ai-boom-re-getty-v-stability-ruling", "node2": "ai-o-stability-ai", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "Stability AI is the defendant" } }, { "node1": "ai-a-stable-diffusion", "node2": "ai-boom-re-getty-v-stability-ruling", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "Getty sued over images used to train Stable Diffusion" } }, { "node1": "ai-a-o1", "node2": "ai-boom-ms-openai-imo-model", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "OpenAI's Olympiad model continued its reasoning-model line" } }, { "node1": "ai-o-openai", "node2": "ai-boom-ms-openai-imo-model", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "OpenAI built the experimental model" } }, { "node1": "ai-boom-ms-openai-imo-model", "node2": "ai-boom-ev-imo-gold-2025", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "claimed gold-level 35 of 42 points" } }, { "node1": "ai-boom-ms-gemini-deep-think-imo", "node2": "ai-boom-ev-imo-gold-2025", "name1": "is won by", "name2": "wins", "direction": 1, "properties": { "Relation": "causation", "Link kind": "contest", "Reason": "officially certified gold-level 35 of 42 points" } }, { "node1": "ai-o-deepmind", "node2": "ai-boom-ms-gemini-deep-think-imo", "name1": "is developed by", "name2": "develops", "direction": 1, "visibility": false, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Google DeepMind built it" } }, { "node1": "ai-boom-ms-openai-imo-model", "node2": "ai-boom-ms-gemini-deep-think-imo", "name1": "competes with", "name2": "competes with", "direction": 0, "properties": { "Relation": "comparison", "Link kind": "rivalry", "Reason": "same score announced two days apart; verification disputed" } }, { "node1": "ai-o-anthropic", "node2": "ai-boom-ms-claude-mythos-preview", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Anthropic built Mythos Preview" } }, { "node1": "ai-boom-ms-claude-mythos-preview", "node2": "ai-boom-ms-claude-fable-5", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "Fable 5 and Mythos 5 are the generally released and partner versions of the Mythos class" } }, { "node1": "ai-boom-pr-claude-code", "node2": "ai-boom-ms-claude-mythos-preview", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "agentic coding and cyber capability line of Anthropic's models" } }, { "node1": "ai-o-anthropic", "node2": "ai-boom-ms-claude-fable-5", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "Anthropic built Fable 5 and Mythos 5" } }, { "node1": "ai-boom-ms-claude-fable-5", "node2": "ai-boom-rg-commerce-anthropic-access", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "regulation", "Reason": "the Commerce directive restricted the models three days after release" } }, { "node1": "ai-o-us-commerce-bis", "node2": "ai-boom-rg-commerce-anthropic-access", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "Commerce Department directive" } }, { "node1": "ai-boom-rg-commerce-anthropic-access", "node2": "ai-o-anthropic", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "Anthropic had to block foreign nationals from the models" } }, { "node1": "ai-boom-na-aljazeera-anthropic-lifted", "node2": "ai-boom-rg-commerce-anthropic-access", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "Al Jazeera reported the lifting" } }, { "node1": "ai-o-deepseek", "node2": "ai-boom-ms-deepseek-v4", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "DeepSeek released V4" } }, { "node1": "ai-boom-ms-deepseek-v3", "node2": "ai-boom-ms-deepseek-v4", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "V4 follows V3 and V3.1 in DeepSeek's line" } }, { "node1": "ai-o-openai", "node2": "ai-boom-ms-gpt-6-astra", "name1": "is developed by", "name2": "develops", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "creation", "Reason": "OpenAI built GPT-6" } }, { "node1": "ai-boom-ms-gpt-5", "node2": "ai-boom-ms-gpt-6-astra", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "GPT-6 followed GPT-5 about a year later" } }, { "node1": "ai-boom-ms-gpt-6-astra", "node2": "ai-a-chatgpt", "name1": "is powered by", "name2": "powers", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "deployment", "Reason": "Astra went to paid ChatGPT users" } }, { "node1": "ai-boom-p-pete-hegseth", "node2": "ai-boom-rg-pentagon-anthropic-risk", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "Hegseth moved to designate Anthropic a supply chain risk" } }, { "node1": "ai-boom-o-pentagon", "node2": "ai-boom-rg-pentagon-anthropic-risk", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "the Department notified Anthropic of the designation" } }, { "node1": "ai-p-donald-trump", "node2": "ai-boom-rg-pentagon-anthropic-risk", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "Trump ordered agencies to stop using Anthropic's technology" } }, { "node1": "ai-boom-rg-pentagon-anthropic-risk", "node2": "ai-o-anthropic", "name1": "is subject to", "name2": "applies to", "direction": 1, "properties": { "Relation": "configuration", "Link kind": "scope", "Reason": "the designation targets Anthropic and its products" } }, { "node1": "ai-boom-na-fortune-pentagon", "node2": "ai-boom-rg-pentagon-anthropic-risk", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "AP report carried by Fortune" } }, { "node1": "ai-boom-na-fortune-pentagon", "node2": "ai-p-dario-amodei", "name1": "is quoted in", "name2": "quotes", "direction": 1, "properties": { "Relation": "information", "Link kind": "quotation", "Reason": "quotes Amodei on challenging the designation in court" } }, { "node1": "ai-a-eu-ai-act", "node2": "ai-boom-rg-eu-digital-omnibus", "name1": "is based on", "name2": "is the basis for", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "lineage", "Reason": "the omnibus amends the Act's schedule and provisions" } }, { "node1": "ai-o-european-commission", "node2": "ai-boom-rg-eu-digital-omnibus", "name1": "is enacted by", "name2": "enacts", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "regulation", "Reason": "the Commission proposed the omnibus on 19 November 2025" } }, { "node1": "ai-boom-na-gibsondunn-omnibus", "node2": "ai-boom-rg-eu-digital-omnibus", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "client alert on the provisional agreement" } }, { "node1": "ai-boom-fr-anthropic-30bn", "node2": "ai-o-anthropic", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "Anthropic is the recipient" } }, { "node1": "ai-o-microsoft", "node2": "ai-boom-fr-anthropic-30bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "previously announced Microsoft investment was included" } }, { "node1": "ai-o-nvidia", "node2": "ai-boom-fr-anthropic-30bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "previously announced Nvidia investment was included" } }, { "node1": "ai-boom-fr-anthropic-30bn", "node2": "ai-boom-fr-anthropic-65bn", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "next Anthropic round three months later" } }, { "node1": "ai-boom-fr-anthropic-65bn", "node2": "ai-o-anthropic", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "Anthropic is the recipient" } }, { "node1": "ai-o-amazon", "node2": "ai-boom-fr-anthropic-65bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Amazon added $5 billion under an earlier commitment" } }, { "node1": "ai-boom-fr-softbank-openai-40bn", "node2": "ai-boom-fr-openai-110bn", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "next OpenAI mega-round" } }, { "node1": "ai-boom-fr-openai-110bn", "node2": "ai-o-openai", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "OpenAI is the recipient" } }, { "node1": "ai-o-amazon", "node2": "ai-boom-fr-openai-110bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Amazon committed $50 billion" } }, { "node1": "ai-o-nvidia", "node2": "ai-boom-fr-openai-110bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "Nvidia committed $30 billion" } }, { "node1": "ai-o-softbank", "node2": "ai-boom-fr-openai-110bn", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "SoftBank committed $30 billion" } }, { "node1": "ai-boom-fr-nvidia-openai-100bn", "node2": "ai-boom-fr-openai-110bn", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "Nvidia's later $30 billion commitment sits in the larger round; the earlier letter of intent may be replaced (sources unclear)" } }, { "node1": "ai-boom-re-musk-sues-openai", "node2": "ai-boom-re-musk-v-openai-verdict", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "the 2024 suit ended in the May 2026 jury verdict" } }, { "node1": "ai-boom-re-musk-v-openai-verdict", "node2": "ai-o-openai", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "the verdict resolved the claims against OpenAI" } }, { "node1": "ai-p-dario-amodei", "node2": "ai-boom-re-pacing-the-frontier", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "signatory" } }, { "node1": "ai-o-anthropic", "node2": "ai-boom-re-pacing-the-frontier", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "Anthropic officially backed the initiative" } }, { "node1": "ai-o-openai", "node2": "ai-boom-re-pacing-the-frontier", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "OpenAI officially backed the initiative" } }, { "node1": "ai-a-pause-letter", "node2": "ai-boom-re-pacing-the-frontier", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "statement", "Reason": "the 2026 letter asks for preparation to slow development rather than a pause now, echoing the 2023 call" } }, { "node1": "ai-boom-ms-sora", "node2": "ai-boom-re-sora-shutdown", "name1": "responds to", "name2": "prompts", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "OpenAI withdrew Sora in 2026" } }, { "node1": "ai-o-openai", "node2": "ai-boom-re-sora-shutdown", "name1": "is issued by", "name2": "issues", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "response", "Reason": "OpenAI announced the discontinuation" } }, { "node1": "ai-boom-na-cbs-sora", "node2": "ai-boom-re-sora-shutdown", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "CBS reported the shutdown" } }, { "node1": "ai-p-liang-wenfeng", "node2": "ai-o-deepseek", "name1": "is founded by", "name2": "founds", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "founding", "Reason": "Liang launched DeepSeek in May 2023" } }, { "node1": "ai-boom-na-cset-action-plan", "node2": "ai-boom-rg-ai-action-plan", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "CSET recap of the plan" } }, { "node1": "ai-boom-na-tbs-oracle", "node2": "ai-boom-mm-oracle-openai", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "Reuters report on the share surge" } }, { "node1": "ai-boom-na-wsj-musk-verdict", "node2": "ai-boom-re-musk-v-openai-verdict", "name1": "is reported in", "name2": "reports on", "direction": 1, "properties": { "Relation": "information", "Link kind": "coverage", "Reason": "WSJ reported the verdict" } }, { "node1": "ai-boom-ms-deepseek-v3", "node2": "ai-scale-ch-h800", "name1": "is used to train", "name2": "is trained on", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "training compute", "Reason": "DeepSeek reported 2.788 million H800 GPU hours for V3" } }, { "node1": "ai-a-gpt-3", "node2": "ai-boom-ms-gpt-35", "name1": "succeeds", "name2": "is succeeded by", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "succession", "Reason": "the GPT line continued with GPT-3.5" } }, { "node1": "ai-o-openai", "node2": "ai-scale-mm-amd-openai-2025", "name1": "is triggered by", "name2": "triggers", "direction": 1, "properties": { "Relation": "causation", "Link kind": "market effect", "Reason": "AMD shares rose after its chip agreement with OpenAI on 6 October 2025" } }, { "node1": "ai-p-sundar-pichai", "node2": "ai-boom-pr-bard", "name1": "is announced by", "name2": "announces", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "announcement", "Reason": "announced Bard in February 2023" } }, { "node1": "ai-p-sundar-pichai", "node2": "ai-boom-ms-gemini-1", "name1": "is announced by", "name2": "announces", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "announcement", "Reason": "announced Gemini 1.0 with Hassabis on 6 December 2023" } }, { "node1": "ai-p-demis-hassabis", "node2": "ai-boom-ms-gemini-1", "name1": "is announced by", "name2": "announces", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "announcement", "Reason": "announced Gemini 1.0 with Pichai on 6 December 2023" } }, { "node1": "ai-p-jensen-huang", "node2": "ai-a-blackwell", "name1": "is announced by", "name2": "announces", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "announcement", "Reason": "presented Blackwell as the processor for the generative-AI era" } }, { "node1": "ai-p-donald-trump", "node2": "ai-a-stargate", "name1": "is announced by", "name2": "announces", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "announcement", "Reason": "announced Stargate at the White House on 21 January 2025" } }, { "node1": "ai-p-sam-altman", "node2": "ai-a-stargate", "name1": "is announced by", "name2": "announces", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "announcement", "Reason": "stood with Trump at the Stargate announcement" } }, { "node1": "ai-p-masayoshi-son", "node2": "ai-a-stargate", "name1": "is announced by", "name2": "announces", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "announcement", "Reason": "chairman of Stargate, at the White House announcement" } }, { "node1": "ai-p-satya-nadella", "node2": "ai-a-microsoft-openai-2019", "name1": "is announced by", "name2": "announces", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "announcement", "Reason": "spoke in the 22 July 2019 announcement" } }, { "node1": "ai-p-satya-nadella", "node2": "ai-boom-fr-microsoft-openai-2023", "name1": "is announced by", "name2": "announces", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "announcement", "Reason": "announced the further multiyear investment in January 2023" } }, { "node1": "ai-p-yoshua-bengio", "node2": "ai-boom-re-cais-statement", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "named among the signatories" } }, { "node1": "ai-p-demis-hassabis", "node2": "ai-boom-re-cais-statement", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "named among the signatories" } }, { "node1": "ai-p-emad-mostaque", "node2": "ai-a-pause-letter", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "named among the signatories" } }, { "node1": "ai-o-wga", "node2": "ai-boom-re-wga-deal", "name1": "is signed by", "name2": "signs", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "statement", "Reason": "the guild reached the agreement with the studios" } }, { "node1": "ai-boom-re-musk-sues-openai", "node2": "ai-p-sam-altman", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "lawsuit", "Reason": "Altman is a defendant" } }, { "node1": "ai-p-sam-altman", "node2": "ai-hist-fr-openai-founding", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "named backer of the founding pledge" } }, { "node1": "ai-p-elon-musk", "node2": "ai-hist-fr-openai-founding", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "named backer of the founding pledge" } }, { "node1": "ai-hist-p-greg-brockman", "node2": "ai-hist-fr-openai-founding", "name1": "receives investment from", "name2": "invests in", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "investment", "Reason": "named backer of the founding pledge" } }, { "node1": "ai-o-deepmind", "node2": "ai-a-alphago-lee-sedol", "name1": "is organised by", "name2": "organises", "direction": 1, "properties": { "Relation": "responsibility", "Link kind": "contest", "Reason": "staged the five-game match and put up the prize" } }, { "node1": "ai-a-altman-ouster", "node2": "ai-p-sam-altman", "name1": "is targeted by", "name2": "targets", "direction": 1, "properties": { "Relation": "causation", "Link kind": "response", "Reason": "the board removed Altman as chief executive" } }, { "node1": "ai-p-garry-kasparov", "node2": "ai-hist-ev-deep-blue-match-1997", "name1": "has competitor", "name2": "competes in", "direction": 1, "properties": { "Relation": "composition", "Link kind": "contest", "Reason": "lost the rematch 3.5 to 2.5" } }, { "node1": "ai-p-lee-sedol", "node2": "ai-a-alphago-lee-sedol", "name1": "has competitor", "name2": "competes in", "direction": 1, "properties": { "Relation": "composition", "Link kind": "contest", "Reason": "lost the match 1 to 4" } }, { "node1": "ai-boom-fr-oracle-openai-300bn", "node2": "ai-boom-o-oracle", "name1": "is funded by", "name2": "funds", "direction": 1, "properties": { "Relation": "dependency", "Link kind": "financing", "Reason": "the compute contract pays Oracle" } }, { "node1": "ai-p-mustafa-suleyman", "node2": "ai-o-microsoft", "name1": "is joined by", "name2": "leaves for", "direction": 1, "properties": { "Relation": "sequence", "Link kind": "move", "Reason": "became chief executive of Microsoft AI in March 2024" } }, { "node1": "ai-question", "node2": "ai-a-turing-1950", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-dartmouth-workshop", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-perceptron", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-ms-eliza", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-ms-shakey", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-ms-xcon", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-pa-hopfield-1982", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-backprop-1986", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-deep-blue", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-lstm", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-ev-darpa-grand-challenge-2005", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-imagenet", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-ev-jeopardy-2011", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-pr-siri", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-ms-google-cat-network", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-alexnet", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-ms-dqn", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-gans", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-bahdanau-attention", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-hist-pa-seq2seq", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-resnet", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-alphago-lee-sedol", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-tpu", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-transformer", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-scale-ms-alphazero", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-bert", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-gpt-2", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-scale-pa-scaling-laws", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-a100", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-gpt-3", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-alphafold-2", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-scale-ms-dall-e", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-scale-pr-github-copilot", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-instructgpt", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-h100", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-scale-pr-midjourney", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-stable-diffusion", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-chatgpt", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-llama", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-gpt-4", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-ms-gemini-1", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-ms-sora", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-blackwell", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-o1", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-nobel-physics-2024", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-nobel-chemistry-2024", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-ms-deepseek-v3", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-a-deepseek-r1", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-pr-claude-code", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-ev-imo-gold-2025", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-ms-gpt-5", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-ms-gemini-3", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-ms-claude-mythos-preview", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-ms-claude-fable-5", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } }, { "node1": "ai-question", "node2": "ai-boom-ms-gpt-6-astra", "name1": "is part of", "name2": "includes", "direction": 1, "visibility": false, "properties": { "Relation": "composition", "Link kind": "topic", "Reason": "a milestone of the history of AI" } } ], "visuals": [ { "nodeType": "Question", "shape": "rectangle", "borderColor": "#FFB300" }, { "nodeType": "Era", "shape": "rectangle", "borderColor": "#3949AB" }, { "nodeType": "Paper", "shape": "rectangle", "borderColor": "#00ACC1" }, { "nodeType": "Model or system", "shape": "circle", "borderColor": "#1E88E5" }, { "nodeType": "Product", "shape": "circle", "borderColor": "#8E24AA" }, { "nodeType": "Chip", "shape": "rectangle", "borderColor": "#5E35B1" }, { "nodeType": "Event", "shape": "circle", "borderColor": "#E53935" }, { "nodeType": "Person", "shape": "circle", "borderColor": "#EC407A" }, { "nodeType": "Organisation", "shape": "rectangle", "borderColor": "#FB8C00" }, { "nodeType": "Reaction", "shape": "hex", "borderColor": "#D81B60" }, { "nodeType": "Regulation", "shape": "rectangle", "borderColor": "#43A047" }, { "nodeType": "Market move", "shape": "hex[6]", "borderColor": "#00897B" }, { "nodeType": "Funding round", "shape": "hex[8]", "borderColor": "#FBC02D" }, { "nodeType": "News article", "shape": "rectangle", "borderColor": "#FDD835" }, { "nodeType": "the grid", "shape": "rectangle", "borderColor": "#C3C8CF" } ], "dataTypes": { "As of": "datetime", "Era order": "int", "Start date": "datetime", "End date": "datetime", "Ongoing": "bool", "Field": "typelist", "Paper kind": "typelist", "System kind": "typelist", "Product kind": "typelist", "Chip kind": "typelist", "Event kind": "typelist", "Country": "typelist", "Organisation kind": "typelist", "Reaction kind": "typelist", "Stance": "typelist", "Jurisdiction": "typelist", "Regulation kind": "typelist", "Status": "typelist", "Move kind": "typelist", "Direction": "typelist", "Deal kind": "typelist", "Publisher": "typelist", "Article kind": "typelist", "Open weights": "bool", "Milestone": "bool", "Venue": "string", "Parameters": "string", "Reported users": "string", "Role": "string", "Ticker": "string", "Headline": "string", "URL": "string", "Date": "datetime", "Founded": "int", "Impact score": "int", "Move USD bn": "dec", "Move percent": "dec", "Amount USD bn": "dec", "Relation": "typelist", "Link kind": "typelist", "Reason": "string" }, "detailView": [ { "nodeType": "*", "own": "*", "others": [] }, { "nodeType": "the grid", "own": [], "others": [] } ], "metadata": { "sourceLink": "https://en.wikipedia.org/wiki/History_of_artificial_intelligence" } }