Overview
- OpenAI announced Tuesday that one of its models produced a claimed solution to the Navier‑Stokes equation in 88 hours using more than 10,000 coordinating agents, about 130 billion tokens and what it described as millions of dollars in compute.
- Researchers Tristan Buckmaster and Levent Alpöge flagged striking similarities between OpenAI’s method and their unpublished approach and asked whether their Codex data were used to train the model.
- OpenAI researchers initially denied using the pair’s prompts or results to direct the run but later said they could not rule out that anonymized data from the duo helped improve the models.
- The Clay Mathematics Institute said it welcomed the announcement but noted that formal validation, publication and prize consideration require community review and can take up to two years before acceptance.
- The effort highlights a shift to very large, multi‑agent AI workflows and raises practical questions about consent, reproducibility and research governance as competition between labs intensifies.