- OpenAI publishes AI-generated proof of Navier-Stokes Millennium Problem with formal Lean verification
- Proof by unreleased model consumed 300B tokens (~$22.5M compute) over one week-long effort
- NYU mathematician alleges OpenAI used his approach after learning of his work in progress
- OpenAI’s Bubeck reportedly asked researcher to remove Anthropic co-author Levent Alpoge’s credit
I’m torn on this one… whether OpenAI stole his approach or his approach was heavily influenced and steered by his ChatGPT sessions and the model was capable of solving it already. Either way, when you’re working with the AI to solve million dollar math problems:
Profile → Settings → Data Controls → “Improve the model for everyone” → OFF