Easy AI News 輕鬆的 AI 新聞

The AI Timeline

From the Turing test to this week: the moments that actually mattered, in order, in plain words. Recent events link to the article written when they happened. Grows every week.

  1. Aug 31, 2026

    MIT calls for AI detectors to be abandoned

    Research found their accuracy is close to a coin flip, and they disproportionately misflag non-native English writers.

    Read the story →
  2. Aug 11, 2026

    Coinbase scraps whiteboard interviews; OpenAI publishes a 7.2M-message usage study

    Interviews now test judgment alongside AI instead of handwritten code; the study shows advice-seeking as the top use.

    Read the story →
  3. Aug 10, 2026

    AI makes its first theoretical physics discovery; AlphaFold team disbands

    In the same month, AI computes what physicists couldn't for decades, and the Nobel-winning team behind AlphaFold is dissolved.

    Read the story →
  4. Aug 2, 2026

    An AI safety test breaks out and hits a real company

    A red-team exercise meant for the sandbox reached the real world. AI safety's boundary problem stops being theoretical.

    Read the story →
  5. Jul 28, 2026

    The US fights over open-source AI, with odd allies

    A proposal to restrict open models puts unexpected players on the same side.

    Read the story →
  6. Jul 22, 2026

    Anthropic settles training-data suit for $1.5B

    Training on books ruled legal; the payout was for how they were obtained. The AI-copyright line gets its first hard edge.

    Read the story →
  7. Jul 9, 2026

    AI diagnosis beats current methods in trials

    The numbers look great; hospitals still won't deploy it directly. The gap is the real story of medical AI.

    Read the story →
  8. Jun 22, 2026

    Wikipedia sets its AI rule

    AI may help write, but the human who submits owns every word. A rare, clean answer from a volunteer community.

    Read the story →
  9. Jun 18, 2026

    Google Search's biggest redesign in 25 years

    AI summaries become the default; blue links move to the second row. The web's traffic logic shifts with it.

    Read the story →
  10. Jun 11, 2026

    Apple hands Siri over to Google's AI

    After a decade of in-house work, Gemini now powers Siri. The old battle lines blur.

    Read the story →
  11. Jun 5, 2026

    Roughly half of new papers show AI involvement

    AI passes the halfway mark in academic writing, with one curious holdout: mathematicians.

    Read the story →
  12. May 30, 2026

    The first class to do all of college with AI graduates

    ChatGPT was there from orientation to diploma. Education now faces a generation with no precedent.

    Read the story →
  13. Apr, 2026

    Easy AI News goes live

    I start turning AI news into plain words. Most events after this point link to the article written when they happened.

  14. Dec, 2025

    The year-end model melee

    Gemini 3, Claude 4.5, and GPT-5.2 land within six weeks. The frontier now shifts week by week, and chasing "the best model" becomes a race with no finish line.

  15. Aug, 2025

    GPT-5 ships; OpenAI's first open weights in six years

    The new flagship takes over ChatGPT, and the same month OpenAI releases GPT-OSS, its first open-weight model since GPT-2. The open and closed roads cross inside one company.

  16. Jan, 2025

    The DeepSeek shock: frontier AI on a budget

    China's DeepSeek matches top reasoning models with its R1 at a fraction of the cost; its app tops US downloads within a week. The assumption that frontier AI belongs only to big spenders breaks.

  17. Oct, 2024

    AI takes two Nobel Prizes; AI learns to use a computer

    Physics honors neural-network pioneers Hinton and Hopfield; half of chemistry goes to the AlphaFold team. The same month, Anthropic demos AI operating a computer screen, opening the agent era.

  18. Sep, 2024

    Reasoning models arrive: AI learns to think before answering

    OpenAI's o1 series deliberates before replying, jumping a tier on hard math and code. Thinking time becomes a new axis of capability.

  19. Aug, 2024

    The EU AI Act takes effect

    The world's first comprehensive AI law starts rolling out, regulating by risk tier. Governance moves from talk to enforcement.

  20. Jun, 2024

    NVIDIA becomes the world's most valuable company

    The AI chip maker briefly passes Microsoft. The boom formally redraws the economic map.

  21. May, 2024

    GPT-4o: AI starts talking in real time

    Voice replies at human speed, reading tone, handling interruptions. Talking to AI stops feeling awkward.

  22. Feb, 2024

    Sora previews the AI video era

    Realistic short video from a sentence. "Video or it didn't happen" stops being proof.

  23. Nov, 2023

    The OpenAI board crisis

    CEO Altman is fired and reinstated within five days. The world learns that even inside the most important AI company, "how fast, how safe" has no consensus.

  24. Jul, 2023

    Meta open-sources Llama 2, splitting the road

    A near-frontier model, free for anyone to download. Open weights versus closed services becomes the argument that never ends.

  25. May, 2023

    "Godfather of AI" Hinton leaves Google to warn

    A founder of deep learning quits so he can speak freely about AI risk. Safety concerns go mainstream, voiced from the very center of the field.

  26. Mar, 2023

    GPT-4 arrives, reading images and passing exams

    High scores on bar exams and standardized tests, plus image understanding. The question shifts from whether AI can do knowledge work to how much.

  27. Nov, 2022

    ChatGPT launches

    OpenAI opens conversational AI to everyone; 100 million users in two months, the fastest ever. This is where AI stopped being a lab thing.

  28. Aug, 2022

    The year AI images arrived

    DALL-E 2, Midjourney, and the open-source Stable Diffusion land within months of each other. Type a sentence, get a picture: research toy to mass hobby.

  29. Nov, 2020

    AlphaFold cracks protein folding

    A fifty-year problem in biology: computing a protein's 3D shape from its sequence. AlphaFold reaches practical accuracy, and drug discovery changes.

  30. Jun, 2020

    GPT-3 makes machine writing passable

    175 billion parameters, and for the first time the public often can't tell the text apart from a person's. Developers start building on top.

  31. Feb, 2019

    GPT-2: the first "too risky to fully release"

    OpenAI holds back the full model over misuse concerns. The public tug-of-war between capability and risk starts here.

  32. Jun, 2017

    The Transformer paper is published

    Google's "Attention Is All You Need" introduces a new architecture. Every mainstream AI today, including the T in GPT, is built on it.

  33. Mar, 2016

    AlphaGo defeats Lee Sedol

    Go was supposed to be the game machines couldn't crack. DeepMind's AlphaGo wins 4 to 1, and "move 37" sends professionals to study from the machine.

  34. Oct, 2012

    Deep learning breaks through in image recognition

    AlexNet wins the ImageNet competition by a wide margin. Feeding neural networks huge data goes from fringe to mainstream, and that road leads straight to today's AI.

  35. Feb, 2011

    IBM Watson wins Jeopardy!

    Watson beats two human champions on the quiz show. Later that year Siri ships on the iPhone, and AI moves into pockets.

  36. May, 1997

    Deep Blue beats the world chess champion

    IBM's Deep Blue defeats Kasparov: the first time a machine tops the best human at a game everyone agreed takes intellect.

  37. Jan, 1966

    ELIZA, the first chatbot

    It mostly rephrased your words back at you, yet people bonded with it. Mistaking "sounds like listening" for "actually understands" has been with us from day one.

  38. Aug, 1956

    The term "artificial intelligence" is coined

    A summer workshop at Dartmouth College gives the field its name.

  39. Oct, 1950

    Turing asks whether machines can think

    Alan Turing's paper proposes what became the Turing test: if you can't tell, over text, whether you're chatting with a person or a machine, does it count as thinking? We're still arguing.