OpenAI produced an unverified Navier-Stokes proof using 10,000 AI agents in 88 hours. It later released 722 manuscripts addressing 372 open math problems. Only 162 included machine-checkable proofs. Three manuscripts were withdrawn after errors were found. The work shows AI excels at normal science within human-built frameworks. The bottleneck is now the shortage of experts to verify machine-made results. Congress is urged to reverse a 40% cut to NSF graduate fellowships. It should also fund paid verification teams for machine-generated science.
What changed
AI research produced results at a pace limited by human expert evaluation time.
What it unlocks
AI can now generate scientific results at industrial scale faster than humans can verify them.
- 10,000 AI agents
- 88 hours to produce the proof
- 722 manuscripts released
- 372 open problems addressed
Sources