Gambling with our lives

Anthropic overnight: Jacob Coxon (pretraining; previously OpenAI technical staff through mid-2026, then Anthropic) said he resigned Tuesday, accusing both labs of racing to self-improving superintelligence and "gambling with our lives." WSJ had the exclusive Tuesday evening; his X thread ran into the night. Evan Hubinger (Anthropic Alignment Science) publicly put greater than 10% odds on AI-caused catastrophe this decade. Not a Dario post — still Anthropic-orbit personnel news.

Also Anthropic: Addy Osmani joined Claude Code. Ex-Google Chrome / Cloud AI developer-experience lead (14 years). No formal title disclosed. Timing late Tuesday / overnight.

OpenAI / Altman overnight follow-through, not a new Millennium claim: after Tuesday's Navier–Stokes post, Altman replied late evening "you know what / let's try" on a room-temperature-superconductor push. Verge and others deepened the Buckmaster/Alpöge credit fight; OpenAI still says no specific user data, won't seek the $1M Clay prize.

Tesla / Musk: replied that few understand how optimized Cybercab is for lowest fully-loaded cost per mile. Separate sniping at Alex Gibney. No new product or EU approval.

Quiet from Hassabis, Karpathy, Huang, and Sutskever. AlphaGenome Atlas stays on yesterday.