OpenAI's 10,000 agents proved a Navier-Stokes variant in 88 hours — sparking plagiarism claims
DeepLearningAI · x · 2026-09-22
OpenAI announced that an unreleased model, run across 10,000 agents over 88 hours, produced a proof showing the Navier-Stokes fluid equations can break down under the right conditions — a problem tied to the $1M Clay Millennium Prize.
The controversy: Mathematicians Tristan Buckmaster and Levent Alpöge had been attacking the same problem with OpenAI Codex, and Buckmaster had discussed it with OpenAI staff. OpenAI published its proof just one day after the pair released proofs of related equations, fueling suspicions the agents saw their unpublished work, possibly via training data.
Lessons for AI developers (DeepLearning.AI): agents can formalize complex proofs at scale; human evaluation remains essential; enterprise zero-data-retention settings matter.
Related event: OpenAI's Navier-Stokes claim sparks debate over math validity and hype(9 posts)→
More from AGI Musings
- Should AI labs publish math results directly or wait for gatekeepers' approval? — TinfoilTricorn · 2026-09-22
- Two years ago AI answered one step; now models work for weeks on a problem — arpitingle · 2026-09-22
- Professor Predicts AI Will Split Universities Into Industrial Research and Boutique Teaching — prof_g · 2026-09-22
- Are there now more builders than buyers for vibe-coded apps? — nikvassev · 2026-09-22
- Agent Tackles Open Ramsey Theory Problem, Drains MacBook Pro Battery Running Optimization — tdhopper · 2026-09-22
- Technical SEO is Dead* — Ahrefs' Patrick Stox on agents and search — gaganghotra_ · 2026-09-22