
Braid · October 7 · 20 min
OpenAI Publishes 722 Math Manuscripts From an Unreleased Model—Not All Verified
0:00 · 722 manuscripts, some checked and some not-20:31
transcript
show notes
OpenAI posted 722 mathematics manuscripts from a model nobody outside the company can run, and some come with machine-checked proofs while others don't. That split sets up the day: who gets to verify, who gets to set the rules for agents, and who gets to say stop to a data center.
- OpenAI's announcement and its post on sharing AI progress in mathematics give the 722 manuscripts and 372 result families, with Lean proofs on many and others unverified; The Verge covers the reaction, and Ethan Mollick's post looks beyond math.
- Jeff Weinstein, Sierra, CNBC via Techmeme and SiliconANGLE cover the Personal Agent Protocol from Meta, Sierra, Stripe, Shopify, Walmart and others, with the spec itself still to be read.
- Axios and Techmeme report Mistral Large 4's specs and staged release, weights due October 27, while Nathan Lambert's Interconnects piece argues GLM-5.3's open release hasn't produced the attacks Anthropic warned about.
- OpenAI's Decisions API video, Strands Decider and Simon Willison's plugin show two small, constrained-choice models arriving the same day.
- CNBC and Techmeme report Finland's stop order to Google by October 23; the FT via Techmeme reports SpaceX's roughly $40 billion debt raise for Nvidia chips; CNBC carries Temasek's warning.
- Amjad Masad, Naval and eigenrobot on AI decompilation, including an unconfirmed Adobe claim.
- Reuters on South Korea's bank-hack statement and Toby Walsh in The Guardian on OpenAI in Australia.
- The Verge via Techmeme on Common Sense Media's rating of ChatGPT for Teens and NBC News on Oversight Board advice.
- Preprints: power draw and covert compute, auditable claims about AI agents and AdvSim2Real.
- AI Engineer talks: Stigg on metering, Resolve AI, Langfuse and Tessl.
links32





