Have a correction or suggestion?
Send it via the form below with a source link.
We review submissions and update regularly so this stays accurate, fair, and useful.
It’s the product of decades of ideas, experiments, breakthroughs, and a lot of hard work and its only just beginning.
This timeline is here to create a visual representation of the journey that has gotten us to where we are today. See how we moved from early theory to real-world tools that write, code, design, reason, and help run modern operations—from the office to the factory floor.
The goal is simple: context you can trust. Each milestone has clear one-liner with sources you may visit to step back to that moment and read more. You’ll find not only the headline releases (GPT-4o, Claude, Gemini, Llama, GPT-5) but also some of the lesser known supporting cast that makes AI valuable—open-source projects, evaluation bodies, policy milestones, and industry adoption in automation and manufacturing. As I started to research it blows my mind how large of an ecosystem of companies and talented individuals have been foundational to this process.
This is a living timeline. AI moves fast, and so will this page. If I’ve missed a key event, date, or global contribution, or if you have a better source, please share it!
Alan Turing’s paper ‘Computing Machinery and Intelligence’ frames the modern AI debate and introduces the Imitation Game.
McCarthy, Minsky, Rochester and Shannon convene the Dartmouth workshop, establishing AI as a field.
Frank Rosenblatt describes the perceptron (early neural network) while John McCarthy creates Lisp, the dominant AI language for decades.
Samuel’s checkers program learns from experience, coining the term ‘machine learning’ and showing self-improving programs in action.
Unimate starts work at GM, marking the dawn of industrial robotics and AI‑adjacent automation on factory floors.
Joseph Weizenbaum’s ELIZA simulates a Rogerian therapist through pattern matching, revealing the ‘ELIZA effect.’
Shakey becomes the first general‑purpose mobile robot that can reason about its actions, integrating perception, planning (STRIPS) and navigation.
UK’s Science Research Council publishes the Lighthill Report criticizing AI progress, leading to major funding cuts and a slowdown in research.
The MYCIN expert system recommends antibiotics for blood infections, showing rule‑based systems can match or exceed human experts in narrow domains.
Hans Moravec’s Stanford Cart navigates a cluttered room without human input—an early milestone in autonomous navigation.
Rumelhart, Hinton & Williams show backpropagation can successfully train deep, multi‑layer neural networks—reviving neural nets.
LeCun’s convolutional network recognizes handwritten digits, deployed for postal ZIP code reading—an early real‑world deep learning win.
IBM’s Deep Blue becomes the first computer to defeat a reigning world chess champion in a match under tournament conditions.
iRobot introduces the Roomba robot vacuum at a $199 price point, putting practical autonomy in millions of homes.
Stanley finishes a 132‑mile autonomous desert course in under 7 hours, jump‑starting modern self‑driving research.
CUDA unlocks general‑purpose computation on GPUs, dramatically accelerating deep learning training.
Watson defeats champions Ken Jennings and Brad Rutter on national TV, showcasing large‑scale QA and NLP.
Krizhevsky, Sutskever & Hinton’s GPU‑trained CNN halves prior error rates, igniting the deep learning revolution in vision.
Goodfellow et al. propose adversarial training between generator and discriminator, enabling photorealistic image synthesis.
DeepMind’s AlphaGo combines deep nets and tree search to beat a world champion years earlier than expected.
Vaswani et al. replace recurrence with self‑attention, laying the foundation for modern LLMs.
Bidirectional Transformer pre‑training delivers SOTA across many NLP tasks with simple fine‑tuning.
Final staged release of GPT‑2 with 1.5B parameters, highlighting safety concerns around synthetic text.
GPT‑3’s massive scale unlocks few‑shot behaviors across tasks, catalyzing LLM apps via API access.
DALL·E shows controllable image synthesis from text prompts, foreshadowing the 2022 image‑model boom.
Copilot brings AI pair‑programming to IDEs, powered by code‑tuned LLMs (Codex).
Open‑weight text‑to‑image model fuels mass experimentation and a vibrant creative ecosystem.
ChatGPT popularizes conversational AI worldwide, reaching ~100M MAUs by Jan 2023 (fastest growth at the time).
GPT‑4 accepts images and text; significantly stronger reasoning and exam performance than GPT‑3.5.
Plugins connect ChatGPT to the web and services; function calling formalizes safe tool use and structured outputs.
Crowdsourced head‑to‑head human preferences yield an Elo‑style leaderboard of LLM quality; later extended to multimodal.
Gemini 1.5 introduces breakthrough long‑context capabilities (128k standard; up to 1M tokens in preview).
MLPerf formalizes gen‑AI inference with large LLM and text‑to‑image tasks for standardized comparisons.
DBRX open‑sources a high‑quality MoE LLM targeting enterprise customization and efficiency.
MLPerf Inference v4.0 expands benchmarks to large LLMs and text‑to‑image, enabling more apples‑to‑apples performance comparisons.
Open‑weight Llama 3 family ships with assistant rollout across Meta products.
Perplexity’s enterprise tier debuts with partnerships and rapid usage growth.
Snowflake unveils Arctic with emphasis on openness and enterprise retrieval/efficiency patterns.
GPT‑4o offers real‑time voice, vision, and text with lower latency/cost; added to ChatGPT free tier.
Pages converts research threads into shareable articles aimed at enterprise and creator workflows.
Apple unveils Apple Intelligence and integrates ChatGPT into Siri and Writing Tools with user approval.
Claude 3.5 Sonnet raises mid‑tier performance and debuts Artifacts: an on‑page workspace for code/docs.
Public leaderboard comparing text/image embedding models across retrieval, clustering, STS and more.
Meta ships a 405B‑parameter open‑weight model alongside 70B/8B variants, expanding open ecosystem options.
The EU AI Act becomes law; obligations phase in through 2025–2027, including GPAI model duties.
o3‑mini debuts as a cost‑efficient reasoning model with adjustable reasoning effort in ChatGPT and API.
Google introduces Gemini 2.0 Pro Experimental with a 2M‑token context window via AI Studio and Vertex AI.
First hybrid reasoning model with toggleable visible ‘extended thinking’ and thinking budgets for developers.
Annual data‑driven report on AI costs, capabilities, incidents, and geopolitics; widely cited for macro trends.
New reasoning lineup; o4‑mini optimized for fast, cost‑efficient reasoning; system card published with benchmarks.
Apple expands Apple Intelligence capabilities across iPhone, iPad, Mac, Apple Watch and Vision Pro.
Perplexity Max launches as a premium tier with early access to features and higher limits for power users.
Perplexity announces Comet, a Chromium‑based research‑centric browser, initially for Max subscribers via invites.
Rules for general‑purpose AI providers start applying; models placed before this date must comply by Aug 2, 2027.
GPT‑5 becomes the default ChatGPT model with user‑selectable Faster/Smarter/Thinking modes; API and enterprise rollout announced.
OpenAI outlines ‘safe‑completions’—a shift toward output‑centric safety training compared to refusal‑based methods.
Perplexity begins rolling out integrated AI video generation to Pro and Max subscribers as a research visualization tool.
Send it via the form below with a source link.
We review submissions & update regularly so this stays accurate, fair, and useful.