Particle.news

OpenAI Publishes Machine‑Checked Navier–Stokes Proposal as Questions Over Data and Credit Persist

Triggering formal verification, the release raises unresolved questions about whether user drafts influenced the model and who should receive authorship credit.

Overview

  • OpenAI says an internal AI system produced a proposed finite‑time singularity solution to the three‑dimensional Navier–Stokes problem and released Lean formalization files to allow machine checking.
  • The company reported using a massively parallel multi‑agent run that generated the proposal in roughly days and completed a Lean formalization in additional hours, claiming thousands of concurrent agents and billions of tokens were involved.
  • Independent review is underway and the Clay Mathematics Institute has opened its standard prize review process but has not made any final determination about prize eligibility or credit.
  • Mathematicians Tristan Buckmaster and Levent Alpöge had parallel unpublished progress and Buckmaster asked whether interactions with OpenAI’s Codex could have influenced the model; OpenAI has denied that those recent Codex prompts trained or were seen by the system.
  • The episode has intensified calls across the field for new norms on verification, human‑level explanation, data‑use policies and authorship when AI produces formal but potentially opaque mathematical results.