OpenAI solves Navier–Stokes Millennium Problem in 88 hours, sparks scooping row
Simon Willison · rss · 2026-09-09
Simon Willison details OpenAI's claim that an unreleased internal model resolved the Navier–Stokes existence and smoothness problem—about 4.9M agent messages and 300B output tokens across attempts, with Lean verification via GPT-6 Astra. The result is shadowed by a dispute: NYU professor Tristan Buckmaster and Anthropic's Levent Alpöge had worked on related problems for nearly a year using Claude and Codex; after word leaked, OpenAI raced ahead and published first. Buckmaster questioned whether his Codex sessions influenced the model; OpenAI denies accessing user data but can't rule out training on de-identified usage. Willison ties this to a broader worry: rumors of an unpublished solution may now trigger millions of dollars of LLM spend to scoop it, and "used to improve models" remains dangerously vague.
More from Fun
- Reddit joke: once the singularity hits, my UBI all goes to tokens — TheNerdosapien · 2026-09-09
- Mathematicians' first reaction to AI solving a century-old problem: credit disputes — Present-Chocolate591 · 2026-09-09
- "10,000 concurrent agents on an impossible task, competitor publishing soon — what could go wrong" — w_hgm · 2026-09-09
- Poke bot hits 1 year and 200M messages; founder-bait banter ensues — aidangch · 2026-09-09
- Indistinguishable AI stunt video sparks call to ban unlabeled AI content accounts — JourneymanChina · 2026-09-09
- Phrack 73 hardcover launch to be hosted exclusively at Unprompted — dyn___ · 2026-09-09