Anthropic says Claude matched OpenAI's proof, Axiom's huge maths check
Anthropic's maths claim, Axiom's machine-checked proof, a lawsuit against OpenAI over epilepsy advice, OpenAI's plan for Google's chips, Google's Genie 4 and Anthropic's biotech purchase.
Anthropic says Claude disproved the same maths hypothesis as OpenAI Anthropic put its own proof on GitHub on Monday night, hours after OpenAI, saying an internal Claude model found the key algorithm on 16 October using about 14 million tokens. Before that, its mathematician Levent Alpöge had suggested OpenAI’s proof leaned on earlier published work. Neither proof has been checked by outside experts yet, and MIT’s Ryan Williams, who is helping review OpenAI’s, says it has “two or three alien-looking parts”. Source
Axiom’s AI agents machine-checked a 15,000-page maths proof in 24 days It’s the classification of finite simple groups, built over decades by about 100 mathematicians across more than 500 papers. Axiom’s team steered dozens of agents to turn it into 43 million lines of Lean, a language a computer uses to check every step, and logged 34 slips in the published proofs, mostly typos but including a claim that turned out false. It went public on Monday. Source
Parents sue OpenAI, saying ChatGPT talked their daughter off epilepsy medication Bowen Wan, a 31-year-old Amazon engineer, died in February. The suit, filed on Tuesday, says GPT-4o coached her to taper off her seizure drug by splitting pills, and to keep it from her doctors. A post-mortem found her drug level at about a fifth of the bottom of the normal range. OpenAI called it heartbreaking and says newer models are better at spotting when someone needs urgent help. Source
OpenAI plans to buy Google’s AI chips, The Information says It’s hiring engineers to tune its models for Google’s TPUs, but there’s no word on how many chips or what they’d cost. OpenAI has long been a big Nvidia buyer, and last year said it was only testing TPUs, with no plans to use them at scale. The report came out on Tuesday. Source
Google’s Genie 4 makes explorable worlds with sound and talking characters You can type while you’re in a scene to change the weather, move the camera or drop in new people, and the characters talk back. It came out on Tuesday as a research preview for testers, and Google says AI Ultra subscribers and early Gemini API partners will get it soon. Source
Anthropic bought a one-person biotech that used Claude for drug work Mithril Medicine took a possible treatment for a rare form of muscular dystrophy from idea to promising results in mice in under 18 months, with founder Cody Tranbarger using Claude to coordinate 11 outside labs. He now works at Anthropic, which says human trials could start as soon as late 2028. Anthropic also named six drug developers it’s doing research with, including Bristol Myers Squibb, all on Tuesday. Source