Sam Altman announced at the ADIPEC opening ceremony in Abu Dhabi that OpenAI’s next GPT model will be the same internal system used to prove part of the Navier-Stokes equations. To solve the complex fluid dynamics problem, the company deployed 10,000 copies of the model that collaborated without a hierarchy for approximately 88 hours. Reports estimate the total compute cost for the mathematical sprint was about $22.5M.
The resulting proof addressed the “forced” version of the problem, which examines fluid behavior under outside pressure, while the more fundamental “unforced” version remains open. Although the solution was verified using the Lean proof-assistant program, the Clay Mathematics Institute has not yet certified the results, and OpenAI stated it will not claim the $1M Millennium Prize reward.