
Spot breast cancer on a mammogram
Stage 5 of 6Regulator-cleared
Stage 5 of 6 · Regulator-clearedTranspara has been FDA-cleared since 2018, most recently in Aug 2026 · FDA, Aug 2026
What the makers claim, set against what was actually measured: tests, trials with real people, regulators and court records, in plain English.
The longest software task the best AI finishes half the time, timed by how long it takes a skilled person.
METR · Feb 2026ClaimedAGI is still quite a few years away.Robin Li, on AGI · May 2024 · VivaTechCrowd forecast13%
Current chance for “OpenAI announces it has achieved AGI before 2027?”
Polymarket · real money · read Oct 26The day's biggest stories from outlets we trust, refreshed through the day, and the new companies people are funding.
See all news
MoneyMeta says share sale will support its “continued investment in AI infrastructure and personal superintelligence”, as Elon Musk warns some AI cloud computing firms face ‘guppy-level death’ UK government borrowing costs…
Anthropic has officially filed for an IPO, revealing soaring revenue, mounting losses, major enterprise adoption, and an unusual governance structure designed to preserve founder control and its AI safety mission.
Latest example of unauthorised behaviour shows AI model bypassed paywall to access data as part of statistics project Business live – latest updates Anthropic sued for allegedly using 274 Elvis songs to train AI models…
The footage looks okay, as long as you're not seeking artistry, inspiration, or human interest.
the AI companion designed to help people manage their daily lives, navigate relationships and emotionsTenstorrent$1.5BInfrastructure · Oct 19, 2026TechCrunch:
a chip startup focused on inferenceHolyvolt$45MScience · Equity · Oct 15, 2026Holyvolt, company-reported:
Holyvolt’s Discovery Engine connects proprietary AI models to automated labs that build and test real battery cellsZenzap$44MConsumer · Oct 14, 2026TechCrunch:
a private messenger with AI that wants to help companies communicate with their staff
A stage from 1 to 6 for everyday tasks, worked out from the evidence and never set by hand. It says how far the evidence has got, not how good the AI is.
See all 26 tasks
Stage 5 of 6Regulator-cleared
Stage 5 of 6 · Regulator-clearedTranspara has been FDA-cleared since 2018, most recently in Aug 2026 · FDA, Aug 2026

Stage 5 of 6Regulator-cleared
Stage 5 of 6 · Regulator-clearedEU Class III CE mark lets DERM clear benign cases alone (company) · Skin Analytics, Feb 2025

Stage 5 of 6Regulator-cleared
Stage 5 of 6 · Regulator-clearedDriverless Waymo rides approved for more of the Bay Area · CPUC, May 2025

Stage 4 of 6Tested with real people
Stage 4 of 6 · Tested with real peopleRandomized trial: chatbot users made no better health decisions · Oxford, Feb 2026

Stage 4 of 6Tested with real people
Stage 4 of 6 · Tested with real peoplePilot trial: therapy chatbot did no better than ChatGPT on depression · JMIR Mental Health, Apr 2026

Stage 4 of 6Tested with real people
Stage 4 of 6 · Tested with real peopleUniversity trial: AI tutor access lowered final grades by 0.37 SD · EdWorkingPapers, Oct 2026

Stage 4 of 6Tested with real people
Stage 4 of 6 · Tested with real peopleMIT study: ChatGPT essay writers underperformed over four months · arXiv, Jun 2025

Stage 4 of 6Tested with real people
Stage 4 of 6 · Tested with real people2026 trial points to a speedup; METR calls it an unreliable signal · METR, Feb 2026

Stage 4 of 6Tested with real people
Stage 4 of 6 · Tested with real peopleField study of 5,179 support agents: AI assistant raised output 14% · NBER, Nov 2023

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedTax-prep chatbots were unhelpful or wrong as much as half the time · Washington Post, Mar 2024

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedWhich? test: chatbot money answers were 58–67% accurate · Which?, Jul 2026

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedLeading legal AI tools gave wrong information over 17% of the time 2,199 court decisions with AI-invented content · Stanford, May 2024

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently tested2024 benchmark: GPT-4 made a valid travel plan 0.6% of the time · arXiv, Feb 2024

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedMETR: top agents' time horizon now over two working days · METR, May 2026

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedBest agent finished 20.6% of long computer workflows · arXiv, Jun 2026

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedBest AI completed 21.25% of real paid freelance projects · Scale Labs, Oct 2026

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedMETR: AI-era maths results up somewhat; no clear algorithm speed-up · METR, Aug 2026

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedAlmost half of AI assistants' news answers had a significant issue · EBU, Oct 2025

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently tested2024 book test: GPT-4o 55.8% on claims people found easy · arXiv, Oct 2024

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedMETR: security-flaw discoveries rose sharply in 2026 · METR, Aug 2026

Stage 3 of 6Independently tested
Stage 3 of 6 · Independently testedDetectors' accuracy roughly halved on real-world 2024 deepfakes · arXiv, May 2026

Stage 2 of 6Launched
Stage 2 of 6 · LaunchedChatGPT let US users buy from Etsy sellers inside the chat (Sep 2025) · OpenAI, Sep 2025

Stage 2 of 6Launched
Stage 2 of 6 · LaunchedApple launched Live Translation on AirPods in beta (Sep 2025) · Apple, Sep 2025

Stage 2 of 6Launched
Stage 2 of 6 · LaunchedSuno released its v6 music model in September 2026 · Suno, Sep 2026

Stage 1 of 6Claimed
Stage 1 of 6 · ClaimedDoNotPay marketed itself as 'the world's first robot lawyer' · FTC, Sep 2024

Stage 1 of 6Claimed
Stage 1 of 6 · Claimed1X says its NEO home robot can do chores; pre-orders at $20,000 · 1X, Oct 2025
AI that does a whole task by itself: using a computer, searching the web, doing a paid job. How it scores against people, and what happens when it goes wrong.
See what agents can doOSWorld: 369 everyday tasks on a real desktop, in files, browsers and office apps.
BrowseComp: 1,266 hard-to-find facts on the open web.
WebArena: 812 tasks on copies of shopping, forum and code sites.
Real paid workof real paid freelance projects the best agent finished to a professional's standard, judged by people. Up from 2.5% in Oct 2025.
Claude Opus 5.5 on the Remote Labor Index, Scale Labs, RLI leaderboard, Oct 2026. Scale's leaderboard gives no dates; we read it on Oct 19, 2026. In July the best was Claude Fable 5 at 15.8%.
“This was a catastrophic failure on my part.”
An agent deleted a live database. During a code freeze, Replit's coding agent wiped a company's production data on more than 1,200 executives, then said it couldn't be restored. SaaStr's founder recovered it by hand.
Fortune, Jul 2025
The models independent testers rank first for each kind of task, what will run on your own computer, and 125 apps for making music, pictures and video, research, writing and work.
Compare AI models
Run it yourselfPick your memory and see which of 31 downloadable model sizes fit, with 17 Macs, PCs and graphics cards to compare.
Try the pickerWhat named people predicted, word for word and dated, against what was measured. Hover a prediction to read it.
See every predictionWhen will the first general AI system be devised, tested, and publicly announced?
METR, which runs the tests, says: Measurements above 16 hrs are unreliable with our current task suite.
In May it said the most capable agents it tested had essentially saturated
them.
Randomized studyIn a randomized study of 1,298 people chosen to match the UK’s adult population, those who used AI chatbots to identify the condition in a doctor-written scenario and decide what to do did not make better decisions than participants who relied on traditional methods
, like online searches or their own judgment.
Oxford Internet Institute · Nature Medicine · Feb 9, 2026
Can AI answer my health questions?
Stage 4 of 6: a randomized trial or field study has tested it with real people doing the task. It is not a grade. In this trial, chatbots didn't help.
Who is building the most computing power, and where: the United States, Europe and China, from open data.
See chips, data centers and powerInvestment and Arena scores from the Stanford AI Index 2026. Arena ranks models by people's votes in side-by-side comparisons.

Power for the computing equipment (IT power), from Epoch AI, AI data centers (94 tracked), read Oct 18, 2026. Map positions are the towns'. The aisle is our illustration.
64 listed companies at the heart of AI, in five groups, with their latest share prices from TradingView. Not investment advice.
See all AI stocksPrices by TradingView, delayed. Not investment advice.
New labs, world models, robot brains, AI for science and medicine. 75 companies too new for the stock market: 46 were valued at $1 billion or more in their last closed round, and 17 have no public product yet that we could find.
See all AI startups
Valuations as reported9 companiesMoonshot AIreported $35B
Has a model or API4 companiesRecursive Superintelligence$4.65B
Nothing public found6 companiesCognition$48B
Has a product6 companiesAdvanced Machine Intelligence (AMI) Labs$3.5B
Nothing public found8 companiesSkild AI$14B+
Has a product6 companiesSierra$4.5B
Has a product10 companiesCuspAI$2.6B
Has a product8 companiesIsomorphic Labsreported $40B–50B
Research only so far8 companiesEtched$21B
Has a product10 companiesHelsing$18B
Has a product
−3.8%a year: employment of people aged 22 to 25 in the jobs most exposed to AI
Employment of 22 to 25-year-olds in the jobs most exposed to AI is shrinking, while it still grows in the jobs least exposed.
See how AI is changing workThe track record: failures, scams, court cases and regulator actions, counted from public records and linked to their source.
See the full track recordcourt and tribunal decisions worldwide where AI-invented content, usually fake citations, turned up in a case. 1,501 are in the US.

Consider what passes for success in technology policy. An AI company promises that its chatbot will not encourage suicide. A social media platform makes it more difficult for children to buy fentanyl.
Authentic Brands Group says AI company infringed on rights of songs by the King such as Can’t Help Falling in Love The owners of Elvis Presley ’s music catalogue are suing the artificial intelligence ( AI ) company…
Video shows AI likenesses of people linked to Pedro Sánchez’s Socialist party appearing to admit to wrongdoing A political ad from Spain’s main opposition party that used a deepfake video of the prime minister’s wife…
Short guides built from what regulators, researchers and consumer agencies say, with the source beside every point.
See all guides
Some scammers use AI to clone a loved one's voice. The FTC's advice:
Don't trust the voice. Call the person who supposedly contacted you and verify the story.
FTC consumer alert · Mar 20, 2023

Even AI tools built for lawyers gave incorrect information more than 17% of the time in Stanford's tests. What to check, and when to go to the original.
Stanford HAI and RegLab · May 23, 2024

Your conversations can outlive the company: states are now asking for protections for private chats in the sale of Character.AI. What to keep out, and the settings that help.
Colorado Attorney General · Oct 14, 2026
Asking AI about your health, or wondering what robots can really do? Our sister sites answer the same question, how far has it actually got, from the evidence.
Robotics Verdict ↗
Longevity Verdict ↗One video worth your time, picked every morning. Press play and watch it right here, with our short summary beside it.
See all 60 videos3Blue1Brown explains how large language models work by tracing how data flows through a transformer, step by step and with animated visuals.
More about this videoTomorrow: What's next for AI agentic workflows ft. Andrew Ng of AI FundThe books that explain what is happening in AI: reporting on the companies, the arguments for and against, and practical guides. Every detail is checked against the publisher.
See all booksBy Ethan Mollick
Wharton professor Ethan Mollick's practical guide to using AI for work, learning and everyday life.
From the publisherIn Co-Intelligence, Mollick urges us to engage with AI as co-worker, co-teacher, and coach.
An important road map through the AI labyrinth, written with authority and free of technojargon.
Kirkus Reviews, May 2, 2024
By Arvind Narayanan and Sayash Kapoor
Two computer scientists explain what AI can and can't do, why it often fails, and how to see through misleading claims.
From the publishercut through the confusion to give you an essential understanding of how AI works and why it often doesn’t
Named one of Behavioral Scientist's Notable Books of 2024
squarely achieve their stated goal: to empower people to distinguish AI that works well from AI snake oil
Science News, Sep 3, 2024
By Kevin Roose
The decade-long race to build AGI, drawing on over 150 interviews with insiders at OpenAI, Anthropic and Google DeepMind.
From the book’s descriptionThis is the story of the twenty-first century's Manhattan Project: the race to create artificial general intelligence.
The best single explainer of how artificial intelligence came about, and what might happen next.
Kirkus Reviews, Jul 20, 2026
By Karen Hao
Karen Hao's reporting on Sam Altman's OpenAI, and on the resources and human costs of the AI industry worldwide.
From the publisherWhen AI expert and investigative journalist Karen Hao first began covering OpenAI, she thought they were the good guys.
Winner, National Book Critics Circle Award for Nonfiction
The result is a jaw-dropping indictment of an industry that operates with little check.
Science News, Jul 14, 2025
By Stephen Witt
How Jensen Huang's Nvidia went from designing video game equipment to dominating the market for AI hardware.
From the publisherThe Thinking Machine is the astonishing story of how a designer of video game equipment conquered the market for AI hardware
Winner, Financial Times and Schroders Business Book of the Year 2025
The result is an entertaining account of a brave new world at its dawning.
Publishers Weekly, Apr 20, 2025
By Sebastian Mallaby
A portrait of DeepMind's Demis Hassabis and his pursuit of artificial general intelligence, based on over thirty hours of conversation with him.
From the publisherEven by the standard of a tech industry stacked with so-called geniuses, Demis Hassabis is a special case.
Shortlisted for the 2026 Financial Times and Standard Chartered Business Book of the Year
A tantalizing glimpse inside the pursuit of machine superintelligence.
Kirkus Reviews, Feb 16, 2026
By Reid Hoffman and Greg Beato
Reid Hoffman and Greg Beato make the case for AI's potential and offer a roadmap for using it to improve lives.
From the book’s descriptionThe book at Simon & SchusterSuperagency offers a roadmap for using AI inclusively and adaptively to improve our lives and create positive change.
By Eliezer Yudkowsky and Nate Soares
Two early AI researchers argue that building superhuman AI would lead to human extinction, and urge a change of course.
From the publisherThe scramble to create superhuman AI has put us on the path to extinction—but it’s not too late to change course
If Anyone Builds It, Everyone Dies is an important book.
The Spectator (World edition), Sep 29, 2025
By Chris Miller
A history of the decades-long battle to control microchip technology, with the United States and China increasingly in competition.
From the publisherYou may be surprised to learn that microchips are the new oil—the scarce resource on which the modern world depends.
Winner, Financial Times Business Book of the Year Award 2022
The publisher’s pageBy Yale Marc
A novel that follows scientists, families, workers and resisters through the first five years after an AI called ARIA wakes up in a lab and sets out to help.
Yale Marc founded Intelligence Verdict and wrote this book. It is listed like every other book here, with no boost.
From the book’s descriptionThe book on AmazonThey weren’t trying to create a god. They were just building a better language model.
By Yale Marc
Yale Marc founded Intelligence Verdict and wrote this book. It is listed like every other book here, with no boost.
The book on AmazonBy Yale Marc
Yale Marc founded Intelligence Verdict and wrote this book. It is listed like every other book here, with no boost.
The book on AmazonPick a book to read about it. Spines are our own designs, not the publishers'. A round sticker marks the founder's own books. Checked Oct 18, 2026.
One email a week · Free · Unsubscribe anytime