Overview
- OpenAI released a proposed proof that an initially smooth fluid can develop a finite-time singularity and published a Lean formalization that machine-checks the logical steps.
- The company says the result came from an internal, unreleased model that coordinated roughly 10,000 AI agents and produced very large outputs over an intensive compute run.
- OpenAI said it will not claim the Millennium Prize, but independent researchers Tristan Buckmaster and Levent Alpöge have publicly disputed credit and suggested OpenAI accelerated after seeing related work while OpenAI denies access to their drafts.
- Mathematicians and the Clay Mathematics Institute must still scrutinize the formal files, confirm that the proof matches the precise prize formulation, and validate any assumptions not present in the machine-checked text.
- The episode highlights a new research model driven by massive multi-agent computation and raises questions about reproducibility, data provenance, authorship and opacity from training techniques that can hide parts of a model’s internal reasoning.