ai_milestones.csv

year,event,category,value_numeric,value_low,value_high,value_unit,estimate_status,source_id,confidence,display_label,notes
1904,Vacuum Tube (Fleming) - enabled electronic switching for computing,Hardware,,,,none,observed,owid,medium,Vacuum Tube,Historical enabling technology; not a training compute point.
1936,Turing Machine (Alan Turing) - defined computability limits,Theoretical Foundation,,,,none,observed,owid,high,Turing Machine,Theoretical milestone; not a training compute point.
1937,Shannon's Thesis - Boolean logic to electrical circuits,Theoretical Foundation,,,,none,observed,owid,high,Shannon,Theoretical milestone; not a training compute point.
1945,ENIAC - first programmable electronic general-purpose computer,Hardware,500,500,500,ops/sec proxy,proxy,kurzweil,medium,ENIAC,Ops/sec proxy shown separately from training FLOPs.
1947,Transistor (Bell Labs),Hardware,,,,none,observed,owid,high,Transistor,Hardware milestone; not a training compute point.
1950,Turing's Computing Machinery and Intelligence,Theoretical Foundation,,,,none,observed,owid,high,Turing Test,Conceptual milestone; not a training compute point.
1956,Dartmouth Conference - artificial intelligence term coined,AI Milestone,,,,none,observed,owid,high,AI Born,Field milestone; not a training compute point.
1957,Perceptron (Rosenblatt),Model Release,1000000,100000,10000000,ops/sec proxy,proxy,kurzweil,low,Perceptron,Proxy for early hardware capability; not training FLOPs.
1958,Integrated Circuit (Kilby),Hardware,,,,none,observed,owid,high,IC,Hardware milestone; not a training compute point.
1959,Arthur Samuel machine learning checkers program,AI Milestone,1000000,100000,10000000,ops/sec proxy,proxy,kurzweil,low,Samuel ML,Proxy for early compute capability.
1965,Moore's Law stated,Hardware,,,,none,observed,owid,high,Moore's Law,Hardware scaling principle; not a training compute point.
1969,ARPANET and Shakey the Robot,Infrastructure;Robotics,1000000,100000,10000000,ops/sec proxy,proxy,kurzweil,low,ARPANET,Proxy for early robotics and network-era compute capability.
1971,Intel 4004 - first commercial microprocessor,Hardware,,,,none,observed,owid,high,Intel 4004,Hardware milestone; not a training compute point.
1973,First AI Winter begins,AI Winter,,,,none,observed,owid,medium,AI Winter I,Funding and expectation milestone.
1986,Backpropagation revival,Model/Architecture,,,,none,observed,owid,high,Backprop,Architecture milestone; not a training compute point.
1987,Second AI Winter begins,AI Winter,,,,none,observed,owid,medium,AI Winter II,Funding and expectation milestone.
1997,Deep Blue defeats Kasparov,AI Milestone,10000000000,10000000000,10000000000,ops/sec proxy,proxy,kurzweil,medium,Deep Blue,Ops/sec proxy; not training FLOPs.
2000,Honda ASIMO,Robotics,,,,none,observed,owid,medium,ASIMO,Robotics milestone.
2002,Roomba mass-market autonomous robot,Robotics,,,,none,observed,owid,medium,Roomba,Robotics adoption milestone.
2006,Deep learning rebrand and deep belief networks,Model/Architecture,,,,none,observed,owid,high,Deep Learning,Architecture milestone.
2006,AWS launch,Infrastructure,,,,none,observed,owid,high,AWS,Infrastructure milestone.
2007,NVIDIA CUDA,Hardware,,,,none,observed,owid,high,CUDA,Hardware/software milestone.
2009,ImageNet dataset,Dataset,,,,none,observed,owid,high,ImageNet,Dataset milestone.
2010,DeepMind founded,AI Milestone,,,,none,observed,owid,high,DeepMind,Institutional milestone.
2011,IBM Watson wins Jeopardy and Siri launches,AI Milestone,,,,none,observed,owid,high,Watson/Siri,AI product milestone.
2012,AlexNet,Model Release,6.00E+17,6.00E+17,6.00E+17,training FLOPs,observed,epoch,high,AlexNet,ImageNet breakthrough.
2012,Google Cat Paper,Model Release,1.00E+16,8.00E+15,1.20E+16,training FLOPs,estimated,epoch,medium,Google Cat,Unsupervised YouTube-scale model estimate.
2013,Word2Vec,Model/Architecture,,,,none,observed,owid,high,Word2Vec,Architecture milestone.
2013,DeepMind DQN,Model Release,1.00E+15,8.00E+14,1.20E+15,training FLOPs,estimated,epoch,medium,DQN,Atari reinforcement learning estimate.
2014,GANs invented,Model/Architecture,,,,none,observed,owid,high,GANs,Architecture milestone.
2015,OpenAI founded TensorFlow open-sourced and ResNet introduced,AI Milestone;Model/Architecture,,,,none,observed,owid,high,OpenAI/ResNet,Institutional and architecture milestone.
2016,AlphaGo defeats Lee Sedol,Model Release,1.00E+18,8.00E+17,1.20E+18,training FLOPs,estimated,epoch,medium,AlphaGo,Training compute estimate.
2017,Transformers paper,Model/Architecture,,,,none,observed,owid,high,Transformers,Architecture milestone.
2018,GPT-1 and BERT,Model Release,3.16E+17,1.00E+17,1.00E+18,training FLOPs,estimated,epoch,medium,GPT-1/BERT,Range midpoint for early transformer models.
2019,GPT-2,Model Release,1.00E+19,8.00E+18,1.20E+19,training FLOPs,estimated,epoch,medium,GPT-2,Estimate.
2020,GPT-3,Model Release,3.14E+23,3.14E+23,3.14E+23,training FLOPs,observed,epoch,high,GPT-3,Published/commonly cited training compute.
2020,AlphaFold,Model Release,1.00E+20,8.00E+19,1.20E+20,training FLOPs,estimated,epoch,medium,AlphaFold,Estimate.
2021,DALL-E and Codex,Generative,1.00E+21,5.00E+20,2.00E+21,training FLOPs,estimated,epoch,low,DALL-E/Codex,Estimate-heavy frontier generation milestone.
2022,ChatGPT based on GPT-3.5,Model Release,3.00E+23,1.00E+23,5.00E+23,training FLOPs,estimated,epoch,medium,ChatGPT,Uses GPT-3.5-era estimate.
2022,Stable Diffusion,Generative,1.00E+20,8.00E+19,1.20E+20,training FLOPs,estimated,epoch,medium,Stable Diffusion,Estimate.
2023,GPT-4,Model Release,2.00E+25,1.00E+25,3.00E+25,training FLOPs,estimated,epoch,medium,GPT-4,Frontier estimate.
2023,Gemini 1.0 Grok-1 and Llama 2,Model Release,3.16E+24,1.00E+24,1.00E+25,training FLOPs,estimated,epoch,low,Gemini/Llama2,Range midpoint across heterogeneous models.
2024,Sora Claude 3 family and o1 reasoning model,Generative;Reasoning/Agentic,3.16E+25,2.00E+25,5.00E+25,training FLOPs,estimated,epoch,low,Sora/Claude3/o1,Cluster estimate.
2024,Gemini 1.5/2.0 and Llama 3.1 405B,Model Release;Regulation,4.00E+25,3.00E+25,5.00E+25,training FLOPs,estimated,epoch,medium,Llama 3.1,Estimate for Llama 3.1 405B and related frontier systems.
2025,Grok-3,Model Release,1.00E+26,1.00E+26,2.00E+26,training FLOPs,projection,source_review_needed,low,Grok-3,Projection pending source review.
2025,o3 Claude 4 family Gemini 2.5/3 and agentic tools,Reasoning/Agentic,2.24E+26,1.00E+26,5.00E+26,training FLOPs,projection,source_review_needed,low,o3/Claude4,Projection over heterogeneous frontier systems.
2025,Quantum error-correction milestones,Quantum/Future Speculative,,,,none,speculative,,low,Quantum,Included as future-facing context; not compute-axis data.
2025,LLaMA 4 Scout,Model Release,4.00E+25,3.00E+25,5.00E+25,training FLOPs,estimated,source_review_needed,low,LLaMA 4 Scout,Estimate pending source review.
2026,Gemini 3.1 Pro,Model Release,2.24E+26,1.00E+26,5.00E+26,training FLOPs,speculative,,speculative,Gemini 3.1,Speculative 2026 projection; visually separated from history.
2026,GPT-5.4,Model Release,1.00E+26,8.00E+25,1.20E+26,training FLOPs,speculative,,speculative,GPT-5.4,Speculative 2026 projection; visually separated from history.
2026,Claude Opus 4.7,Reasoning/Agentic,1.00E+26,8.00E+25,1.20E+26,training FLOPs,speculative,,speculative,Claude 4.7,Speculative 2026 projection; visually separated from history.
2026,Claude Mythos Preview,Reasoning/Agentic,2.00E+26,1.50E+26,3.00E+26,training FLOPs,speculative,,speculative,Claude Mythos,Speculative invite-only preview.
2026,GPT Spud / GPT-5.5,Model Release,5.00E+27,3.00E+27,6.00E+27,training FLOPs,speculative,,speculative,GPT Spud,Speculative pre-release projection.
2026,Agentic AI and recursive self-improvement loops emerge at scale,Reasoning/Agentic,1.00E+27,5.00E+26,2.00E+27,training FLOPs,speculative,,speculative,Agentic AI,Speculative systems-level projection.
2026,Tesla Optimus production pivot and world models,Robotics;Generative,,,,none,speculative,,speculative,Optimus,Speculative robotics context; not compute-axis data.
2026,Omni-modal single models with recursive learning,Speculative,1.00E+28,5.00E+27,2.00E+28,training FLOPs,speculative,,speculative,Omni-modal,Speculative upper frontier projection.