Spaghetti in the equations

OpenAI tonight (~18:20 London): a next-gen internal model "significantly more capable than GPT-6 Astra," run as about 10,000 coordinating agents for about 88 hours, produced an analytical proof plus Lean formalization that Navier–Stokes dynamics can develop a finite-time singularity (they describe an inward-spiraling, spaghetti-like vortex). They frame it as a solution to the Millennium Prize problem. Altman: one of the most amazing moments in OpenAI history this past week. Same thread congratulates Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic) on related fluid-dynamics work; OpenAI says it did not see their materials before public release and used no specific user data.

Parallel: Buckmaster's public statement alleges OpenAI's Sébastien Bubeck pressed for coordinated announcements and to drop Alpöge from authorship. Bubeck called that false and inflammatory. Independent write-ups still flag forced vs unforced Clay wording — treat "solved" as OpenAI's claim, not settled community consensus.

DeepMind / Hassabis today: AlphaGenome Atlas — precomputed predictions for all ~9 billion possible single-letter DNA variants in the human genome (~1 PB), free academic portal + API, plus an AlphaGenome Variant Impact score mixing AlphaGenome and AlphaMissense. Hassabis tied it to AlphaFold's protein map. Research use only; not clinical.

Tesla / Musk: on an Austin Cybercab rider hoping for Europe, Musk replied "Hopefully soon in Europe too." No new EU approval in this window.

Quiet from Karpathy, Amodei (beyond the Buckmaster/Alpöge thread), Huang, and Sutskever.