Simon Willison on the OpenAI Navier-Stokes Millennium Prize story and the data-use question
ivan_bezdomny · x · 2026-09-09
Simon Willison weighs in on OpenAI's claim that an unreleased model produced a resolution to the Navier–Stokes existence and smoothness problem — one of the seven $1M Millennium Prize Problems — a result now clouded by an accusation from NYU mathematician Tristan Buckmaster.
Key points:
- Buckmaster and Levent Alpöge (now at Anthropic) worked on the problem for nearly a year, using Claude and Codex (mainly GPT-5.6 Sol), breaking through on August 15.
- After news of their work reached OpenAI, Buckmaster pressed the company on timing; OpenAI admitted its first prompt was sent only after learning of their work.
- Asked whether the model was trained on, or had access to, the drafts stored in their Codex sessions, OpenAI said the model did not look up user data but gave no answer on training.
- Willison argues the episode exposes how undefined "using your data to improve model performance" still is.
More from Safety
- FDA's GenAI medical device paper exposes the accountability gap in AI clinical practice — jonc101x · 2026-09-09
- US and China reportedly preparing dedicated AI safety talks amid intense AI race — VraserX · 2026-09-09
- Researcher Joe Benton joins METR, warning industry may impose unprecedented global risk — saprmarks · 2026-09-09
- No official US-China AI risk channel exists; a reporting agreement is far more likely than monitoring — zephyr_z9 · 2026-09-09
- micro1 moves to acquire Spirit's data, opens channel to reassure ex-employees — Exp_Mark · 2026-09-09
- Meta details Muse agent safety: sandboxed harness, Sentinel gatekeeper, user-held encryption — unixterminal · 2026-09-09