New Scientist NL · Science

AI Claims Solution to Navier-Stokes Problem, Discussion Arises

OpenAI claims an AI model has solved the Navier-Stokes problem, one of the most difficult mathematical riddles. This followed shortly after partial solutions from other researchers, causing discussion about its genesis.

The Navier-Stokes problem, a Millennium Prize Problem, concerns fluid flow. The question is whether the 200-year-old equations always remain logical or experience 'blow-ups'.

OpenAI used 1000 AI agents for the Euler problem, and 10.000 agents extended this to Navier-Stokes. This took a total of 61 hours.

The cost for a customer would be 15 million dollars. The AI model was more powerful than GPT-6 Astra.

The announcement followed that of Tristan Buckmaster and Levent Alpöge, who solved partial problems. They demonstrated blow-ups in Euler equations.

Buckmaster and Alpöge received help from LLMs. He asked OpenAI for clarification on their methodology, which suspiciously resembled theirs.

OpenAI denied having used the work of Buckmaster and Alpöge. They stated that their Euler proof was different and that no one had access to the work in progress.

AI proofs are difficult to understand and offer less insight. The rapid results leave little time for discussion and evaluation, also for the prize money.

AI-samenvatting op basis van de bron.