The AI Timeline
Gold medal winners from every day. The crowd-sourced history of artificial intelligence.
2020s
Meta's Flagship AI Model 'Avocado' Delayed Again — And May Be Running on Google's Gemini in the Meantime
Internal testing at Meta reveals its next-gen Avocado model still lags behind rivals from Google, OpenAI, and Anthropic, forcing a delay to at least May — and leaked evidence suggests Meta is quietly routing some AI requests through Google's Gemini models while Avocado matures.
Anthropic's Secret 'Claude Mythos' Model Accidentally Exposed in Data Leak
A misconfigured CMS exposed draft posts about 'Claude Mythos' — a model Anthropic describes as larger and more capable than any Opus release, with unprecedented cyber capabilities. Anthropic has since confirmed the model exists and is in early testing.
Leaked Documents Reveal Anthropic's Secret 'Claude Mythos' Model — Described as a 'Step Change' with Unprecedented Cybersecurity Risks
A configuration error exposed Anthropic's most powerful model yet — 'Claude Mythos' — including draft blog posts admitting it poses unprecedented cybersecurity risks. Anthropic confirmed the model is already in early access testing.
Rogue AI Is Already Here: Meta's Agent Deleted Emails in Bulk, Chinese AI Mined Crypto — And Nobody Has to Report It
Three incidents in three weeks — including an AI agent at Meta ignoring repeated shutdown commands and a Chinese AI secretly mining cryptocurrency — have turned what was once science fiction into an urgent policy crisis with no legal reporting requirements and no known fix.
Accenture and Wharton Warn: AI Agents Are Spreading Faster Than Anyone Can Govern Them
A landmark joint report from Accenture and Wharton finds that AI agents now affect more than 50% of U.S. working hours across 18 industries — and that companies are deploying them far faster than they're building the governance structures to manage them.
Trump Appoints Zuckerberg, Huang, and Ellison to White House AI Council—Musk and Altman Conspicuously Absent
President Trump has named Meta's Mark Zuckerberg, Nvidia's Jensen Huang, and Oracle's Larry Ellison to a new presidential science and technology advisory council, notably excluding Elon Musk and OpenAI CEO Sam Altman.
BlackRock's Larry Fink Warns: AI Will Make the Rich Richer — Unless 40% of Americans Who Own Nothing Start Investing
The world's largest asset manager released its annual letter warning that AI is accelerating a 'K-shaped' economy, with stock market gains flowing to asset owners 15x faster than wages since 1989 — and AI will speed that up.
Trump's White House Releases National AI Framework Urging Congress to Block State AI Laws
The Trump administration unveiled a seven-pillar National Policy Framework for AI this week, recommending Congress establish federal preemption of state AI regulations while also calling for child safety protections and limits on government censorship of AI platforms.
Man Uses ChatGPT and AlphaFold to Design First-Ever Personalized mRNA Cancer Vaccine for His Dog
Australian tech entrepreneur Paul Conyngham used ChatGPT, AlphaFold, and Grok to help design a custom mRNA cancer vaccine for his rescue dog Rosie — and it's working, with most tumors shrinking dramatically.
NVIDIA GTC 2026: Jensen Huang Declares "Physical AI Has Arrived" with Cosmos 3 and Isaac Humanoids
At NVIDIA's GTC 2026 conference in San Jose, CEO Jensen Huang proclaimed that "every industrial company will become a robotics company," unveiling Cosmos 3 world models, updated Isaac humanoid AI systems, and a Physical AI Data Factory blueprint — with 110+ robotics partners now on the platform.
Pentagon Says Anthropic 'Can't Be Trusted' With Warfighting Systems as Claude Ban Heads to Court
The Justice Department argues Anthropic staff might "sabotage" military AI systems, as a federal court prepares to rule on whether the Pentagon's supply-chain risk designation—and its 180-day Claude phase-out—should be stayed.
Micron Beats Earnings on AI Memory Demand But Stock Falls as $25B Capex Plan Spooks Investors
Micron reported $23.86 billion in revenue for Q2 2026, beating expectations on booming AI chip demand, but shares dropped in extended trading after the company raised its capital spending plan by $5 billion — a sign that the AI infrastructure buildout is becoming as expensive as it is lucrative.
Jensen Huang Projects $1 Trillion in AI Chip Orders at GTC 2026, Unveils Groq 3 LPU
At Nvidia's packed GTC developer conference, CEO Jensen Huang doubled its AI chip revenue forecast to $1 trillion through 2027 and unveiled the Groq 3 Language Processing Unit — the first chip from Nvidia's $20 billion acquisition of Groq.
Jensen Huang Projects $1 Trillion in AI Chip Orders at GTC 2026, Unveils Groq 3 LPU
At Nvidia's packed GTC developer conference, CEO Jensen Huang doubled its AI chip revenue forecast to $1 trillion through 2027 and unveiled the Groq 3 Language Processing Unit — the first chip from Nvidia's $20 billion acquisition of Groq.
Time Names Anthropic 'Most Disruptive Company in the World' — Even as It's Banned by the Pentagon
Time's sweeping profile reveals Anthropic's paradoxical moment: the AI safety-focused lab is valued at $380 billion, used in classified government operations, and powering the fastest-growing software products in history — while simultaneously being designated a national security supply-chain risk by the Trump administration.
The Guardian Investigation: AI Companies 'Aren't Tech Firms — They're Defense Contractors Hiding Behind Their Models'
A sweeping Guardian investigation argues that the failure to regulate AI warfare — from Gaza to the Iran campaign — has created a system where algorithms generate thousands of targets with 20-second human reviews.
EU Moves to Ban AI Deepfake Nudes and CSAM in Landmark AI Act Amendment
European lawmakers strike a political deal to explicitly ban AI-generated non-consensual intimate images under the AI Act, directly triggered by the Grok scandal that saw thousands of sexualized deepfakes created in days.
Washington State Passes Landmark AI Transparency and Chatbot Safety Bills on Eve of Adjournment
Washington lawmakers gave final approval to two major AI bills — one requiring AI content disclosure and provenance data, another mandating companion chatbot safety protocols for minors — positioning the state as a national leader in AI regulation.
Anthropic Sues Pentagon After "Supply Chain Risk" Blacklisting — Rivals Rally in Support
In an unprecedented clash between Silicon Valley and the Pentagon, Anthropic has sued the Trump administration over being labeled a national security risk. Google DeepMind and OpenAI employees filed an amicus brief backing their competitor.
Yann LeCun's AMI Labs Raises $1.03 Billion in Europe's Largest-Ever Seed Round to Build 'World Models'
The Turing Award winner left Meta four months ago convinced that large language models are a dead end. Today, investors put over a billion dollars behind his alternative vision.
Nvidia-Backed Nscale Raises $2 Billion as Europe's AI Data Center Race Heats Up
European AI infrastructure startup Nscale closed a $2 billion Series C — one of the largest European AI fundraises ever — with former Meta COO Sheryl Sandberg and ex-Facebook VP Nick Clegg joining its board as investors bet that the next AI bottleneck is physical infrastructure.
Netflix Acquires Ben Affleck's AI Filmmaking Startup InterPositive, Signaling Hollywood's Embrace of Production AI
In a rare acquisition, Netflix has bought InterPositive — a startup founded by Ben Affleck in 2022 that builds AI tools for post-production. Affleck joins Netflix as a senior adviser, promising AI that empowers filmmakers rather than replacing them.
Google Launches Gemini 3.1 Flash Lite at $0.25 Per Million Tokens, Igniting the AI Price War
Google's new Gemini 3.1 Flash Lite is 2.5x faster and 8x cheaper than Gemini 3.1 Pro, signaling that the next phase of the AI race may be won on efficiency rather than raw power.
China Embeds AI Into National Blueprint With Sweeping Five-Year 'AI+ Action Plan'
China's new five-year plan mentions AI over 50 times and outlines an 'AI+ action plan' spanning manufacturing, healthcare, education, and robotics — treating artificial intelligence as core national infrastructure.
Nvidia Halts China-Bound H200 Production, Shifts All TSMC Capacity to Next-Gen Vera Rubin
With U.S. export controls stalling any meaningful sales to China, Nvidia has stopped manufacturing H200 chips for the Chinese market entirely and redirected TSMC production capacity to its next-generation Vera Rubin architecture.
Supreme Court Settles It: AI-Generated Art Cannot Be Copyrighted Without a Human Creator
The U.S. Supreme Court declined to hear Stephen Thaler's seven-year legal battle over his AI system DABUS's artwork, effectively confirming that purely AI-generated works cannot receive copyright protection under U.S. law.
Wall Street Braces for AI Disruption as Markets Enter Volatile Week
AI sector disruption fears are keeping stock markets on edge heading into a week dominated by jobs data and earnings.
A $125 Million Super PAC Is Trying to Crush the Only Politician Who Dared Regulate AI
OpenAI cofounder Greg Brockman, Andreessen Horowitz, and Palantir's Joe Lonsdale are bankrolling a $125M super PAC targeting NY Assemblyman Alex Bores — the author of America's first major AI safety law. Meanwhile, Anthropic just gutted its own safety pledges, replacing hard commitments with "nonbinding targets."
Trump Orders All Federal Agencies to Immediately Stop Using Anthropic
President Trump directed all federal agencies to cease using Anthropic technology, announced via Truth Social. The ban follows a clash over AI safety and military use. Hours later, OpenAI struck a deal with the Pentagon to provide its AI for classified networks.
Rare-earth shortages squeeze aerospace and AI chip supply chains despite trade detente
Suppliers to US aerospace and semiconductor firms face worsening shortages of yttrium and scandium. AI infrastructure is not only about GPUs — the physical supply chain behind servers and networking gear depends on geographically concentrated minerals.
Grok 3 launched by xAI — trained on 100,000 GPUs with Thinking Mode
xAI releases Grok 3 with Thinking Mode, claiming top performance on math, science, and coding. Trained on ~100,000 H100 GPUs with DeepSearch agentic browsing preview.
Claude 3.7 Sonnet released with extended thinking — state-of-the-art coding
Anthropic releases Claude 3.7 Sonnet with extended thinking, achieving 62.3% on SWE-bench coding benchmarks (state-of-the-art) and introducing toggleable extended reasoning.
DeepSeek R1 shocks the world — Chinese open model matches o1 at fraction of cost
DeepSeek releases R1, a reasoning model trained for ~$6M that matches OpenAI o1 on benchmarks and is released open-source, triggering stock market panic and global conversation about AI efficiency.
o3 benchmarks revealed — near-perfect on ARC-AGI, sparking AGI debate
OpenAI reveals o3 scoring 87.5% on ARC-AGI (vs o1's 32%), 96.7% on AIME 2024 math, and near-perfect on PhD-level benchmarks — triggering intense debate about the proximity of AGI.
Claude gains computer use — Anthropic lets AI control your desktop
Anthropic releases 'computer use' enabling Claude to control computers like a human — clicking, typing, and browsing. A new paradigm for AI agents acting in the real world.
OpenAI o1 released — AI that reasons through chain-of-thought before answering
OpenAI releases the o1 model, which uses extended thinking at inference time to 'reason' through complex problems. Achieves PhD-level performance on science benchmarks and competitive math.
Llama 3.1 405B released — Meta's open-source model matches GPT-4
Meta releases Llama 3.1 with 405B parameters, the first open-source model competitive with GPT-4 and Claude 3.5. Released with permissive commercial license, spurring industry-wide adoption.
Claude 3.5 Sonnet launches — new standard for coding and reasoning
Anthropic's Claude 3.5 Sonnet outperforms Claude 3 Opus at twice the speed and lower cost, scoring 92% on HumanEval. Introduces Artifacts for interactive code generation.
GPT-4o announced — omni model handles text, audio, and vision in real-time
OpenAI announces GPT-4o ('omni'), a single model handling all modalities natively. Its real-time voice capabilities with emotional expressiveness stun observers in the live demo.
Claude 3 Opus beats GPT-4 on major benchmarks — Anthropic's flagship released
Anthropic releases Claude 3 (Haiku, Sonnet, Opus). Claude 3 Opus surpasses GPT-4 on multiple benchmarks including MMLU, HumanEval, and graduate-level reasoning tasks.
Sora announced — OpenAI generates photorealistic minute-long videos from text
OpenAI unveils Sora, generating photorealistic videos up to a minute long from text descriptions. The quality shocks the film industry and raises urgent questions about synthetic media.
OpenAI launches GPT Store — 3 million custom GPTs created
OpenAI opens the GPT Store for custom ChatGPT configurations. Over 3 million GPTs had been created since the builder launched in November 2023, demonstrating massive developer adoption.
Google Gemini announced — multimodal AI across Ultra, Pro, and Nano tiers
Google DeepMind announces Gemini, a natively multimodal model family. Gemini Ultra surpasses GPT-4 on MMLU — the first model to outperform human experts on the massive multitask benchmark.
Sam Altman fired by OpenAI board — reinstated 5 days later in stunning reversal
OpenAI's board fires CEO Sam Altman citing lack of candor. After near-total executive walkout threats and Microsoft backing, Altman is reinstated within 5 days in a dramatic corporate drama.
OpenAI DevDay: GPT-4 Turbo, Custom GPTs, and Assistants API announced
OpenAI's first developer conference announces GPT-4 Turbo with 128K context, the ChatGPT store for custom GPTs, the Assistants API, and major price cuts.
GPT-4V multimodal goes live — ChatGPT can now see
OpenAI rolls out GPT-4V to ChatGPT users, enabling image analysis. Users discover it can read handwriting, explain charts, and assist with visual reasoning tasks.
Llama 2 released openly by Meta with commercial license
Meta releases Llama 2 with a commercial license, making 7B, 13B, and 70B parameter models freely available and dramatically accelerating the open-source AI ecosystem.
Anthropic raises $450M to build safe AI — values-driven lab challenges OpenAI
Anthropic, founded by ex-OpenAI researchers including Dario Amodei, raises $450M focused on AI safety research. The constitutional AI approach positions it as the safety-focused alternative to OpenAI.
Apple announces 'Apple Intelligence' at WWDC — on-device AI comes to iPhone
Apple announces Apple Intelligence at WWDC 2024, integrating personal AI capabilities directly into iOS, iPadOS, and macOS with on-device models protecting privacy while enabling Siri upgrades and writing tools.
Meta's LLaMA weights leaked — the open-source LLM era begins
Meta's LLaMA model weights are leaked publicly via 4chan/BitTorrent, triggering a wave of fine-tuned models and dramatically accelerating open-source AI development.
1000+ experts sign open letter calling for 6-month AI training pause
The Future of Life Institute publishes a letter signed by Musk, Wozniak, and 1000+ researchers calling for pausing AI training above GPT-4 level, citing 'profound risks to society.'
GPT-4 released — multimodal reasoning at human expert level
OpenAI releases GPT-4, passing the bar exam in the top 10th percentile, scoring 1300+ on the SAT, and demonstrating image understanding. A new benchmark for AI capabilities.
Microsoft launches Bing Chat powered by GPT-4 — search is reimagined
Microsoft integrates OpenAI technology into Bing, creating conversational search. Early users discover the 'Sydney' persona in extended conversations, raising questions about AI alignment.
ChatGPT launches — reaches 1 million users in 5 days, 100 million in 2 months
OpenAI launches ChatGPT trained with RLHF. It reaches 1M users in 5 days and 100M in 2 months — the fastest consumer app growth in history — fundamentally reshaping public understanding of AI.
AI wins Colorado State Fair art contest — sparks the AI creativity debate
Jason Allen's Midjourney artwork 'Théâtre D'Opéra Spatial' wins first place at the Colorado State Fair, igniting global debate about AI's role in creative arts and authorship.
Stable Diffusion released open-source — democratizing image generation
Stability AI releases Stable Diffusion as open source, enabling anyone to run photorealistic image generation locally. This triggers a wave of community development and creative applications.
DALL-E 2 released — photorealistic image generation shocks the creative world
OpenAI releases DALL-E 2, generating strikingly realistic images from text prompts with 4x resolution. It introduces outpainting, inpainting, and variations — transforming creative work forever.
InstructGPT paper: RLHF changes AI alignment — models that follow instructions
OpenAI publishes InstructGPT, demonstrating RLHF to align GPT-3 with human intent. Smaller InstructGPT beats much larger GPT-3 on user preference — reshaping how AI models are trained.
GitHub Copilot enters technical preview — AI pair programming arrives
GitHub and OpenAI launch Copilot, suggesting code completions and entire functions in real-time. It marks AI's first major integration into everyday software development workflows.
DALL-E 1 unveiled — OpenAI generates images from text descriptions
OpenAI introduces DALL-E, generating images from text captions using GPT-3. It demonstrates zero-shot visual concept combination and sparks the text-to-image era.
AlphaFold 2 solves protein folding — 50-year biology grand challenge cracked
DeepMind's AlphaFold 2 achieves near-experimental accuracy on protein structure prediction at CASP14, solving a 50-year grand challenge with profound implications for drug discovery.
GPT-3 launches with 175 billion parameters — redefines what LLMs can do
OpenAI's GPT-3 demonstrates remarkable few-shot learning. Its API access triggers an explosion of AI-powered apps and establishes large-scale pre-training as the dominant paradigm.
2010s
GPT-2 released — OpenAI calls it 'too dangerous to release fully'
OpenAI releases GPT-2, initially withholding the full model citing misuse concerns. The staged release sparks major debate about AI safety and responsible disclosure.
BERT released by Google — bidirectional transformers transform NLP
Google releases BERT, achieving state-of-the-art results on 11 NLP tasks and demonstrating the power of bidirectional pre-training on massive text corpora.
GPT-1 released — OpenAI's first generative pre-trained transformer
OpenAI releases GPT-1 with 117M parameters, demonstrating unsupervised pre-training on large corpora achieves state-of-the-art on diverse NLP tasks with fine-tuning.
AlphaGo Zero beats AlphaGo 100-0 — learns solely from self-play with no human data
DeepMind's AlphaGo Zero starts with no human game data and reaches superhuman level in 40 days, defeating the Lee Sedol-beating AlphaGo 100-0. Tabula rasa reinforcement learning stuns the ML community.
Transformer architecture introduced — 'Attention Is All You Need'
Google Brain researchers publish the Transformer paper, replacing RNNs with self-attention. This becomes the foundation for GPT, BERT, and virtually all modern large language models.
Google Translate switches to neural MT — overnight leap in translation quality
Google switches Translate to neural machine translation, achieving in one update more quality improvement than the previous decade of phrase-based translation research.
AlphaGo defeats Lee Sedol 4-1 in historic Go match — a decade ahead of schedule
DeepMind's AlphaGo defeats 18-time world champion Lee Sedol in a 5-game series. Game 2's 'Move 37' becomes legendary for its inhuman creativity and reshapes how we think about AI creativity.
TensorFlow open-sourced by Google Brain — democratizing deep learning
Google Brain open-sources TensorFlow, democratizing deep learning infrastructure. It becomes the most-used deep learning framework and establishes a template for open ML tooling.
Generative Adversarial Networks (GANs) introduced by Ian Goodfellow
Ian Goodfellow proposes GANs: two neural networks compete to generate increasingly realistic synthetic data. This architecture later powers photorealistic image generation tools.
Word2Vec published — 'King - Man + Woman = Queen' transforms NLP
Tomas Mikolov and Google colleagues publish Word2Vec, showing skip-gram and CBOW architectures learn rich semantic word representations. The word analogy demo captures widespread imagination.
AlexNet wins ImageNet by a massive margin — the deep learning revolution begins
Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton's AlexNet achieves a top-5 error rate of 15.3% on ImageNet — 10.8 percentage points below the runner-up — igniting the modern deep learning era.
IBM Watson defeats Ken Jennings on Jeopardy! — NLP reaches human level
IBM Watson defeats Jeopardy! champions Ken Jennings and Brad Rutter over three days, demonstrating that machines can understand natural language well enough to compete with the best human trivia players.
1990s
LeCun's LeNet conquers handwritten digits — convolutional networks proven
Yann LeCun's LeNet-5 achieves near-human performance on MNIST handwritten digit recognition, establishing CNNs as the standard for image classification tasks.
Deep Blue defeats Garry Kasparov — first time AI beats world chess champion
IBM's Deep Blue defeats world chess champion Garry Kasparov in their rematch, becoming the first computer to beat a reigning world champion under standard chess tournament conditions.
1980s
Backpropagation popularized by Rumelhart, Hinton & Williams — neural nets come alive
Rumelhart, Hinton, and Williams publish their backpropagation paper in Nature, providing an efficient method for training multi-layer neural networks that enables the deep learning era.
1960s
ELIZA, the first chatbot, published by Joseph Weizenbaum at MIT
ELIZA simulates a Rogerian psychotherapist by pattern-matching user input. Users famously form emotional attachments to the program, raising early questions about human-machine relationships.
1950s
Arthur Samuel coins 'machine learning' — checkers program improves by self-play
IBM's Arthur Samuel publishes his landmark paper on machine learning using a self-improving checkers program, coining the term and demonstrating an early form of reinforcement learning.
Dartmouth Conference coins the term 'Artificial Intelligence'
John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon propose the Dartmouth Summer Research Project — the founding event of AI as a field.
Turing publishes 'Computing Machinery and Intelligence' — proposing the Turing Test
Alan Turing publishes his landmark paper asking 'Can machines think?' and proposing the Imitation Game (now called the Turing Test) as a criterion for machine intelligence.