OpenAI says its AI found a solution to the Navier-Stokes problem

In short: OpenAI says an AI system produced a solution to a famous fluid equation problem, but another mathematician disputes how credit and timing were handled.

What happened

OpenAI said it has an AI-generated solution related to the Navier-Stokes equations, a set of equations that describe how fluids like water and air move. The Navier-Stokes problem is one of the Clay Millennium Prize problems, a list of seven hard math challenges with a $1 million prize for each.

OpenAI researcher Sebastien Bubeck said the company started training a new math-focused AI model on August 28. After hearing rumors that Anthropic might be making progress, OpenAI put more resources on the problem. Bubeck said more than a thousand AI “agents” worked on it for over 50 hours, and the result was “Lean-formalized,” meaning written in a proof-checking programming language (like turning a hand-written math solution into a step by step recipe a computer can verify).

OpenAI also said the work used far more computing power than its past math efforts. Mark Chen, OpenAI’s head of research, said the computing cost was “in the millions of dollars.”

The announcement was quickly followed by a public dispute. Mathematician Tristan Buckmaster of NYU, working with Anthropic researcher Levent Alpöge, posted documents describing their own advances in a related area. Buckmaster also claimed OpenAI ramped up after learning about their progress and made proposals about how results would be published and credited. OpenAI denied seeing their work before it was public, and said it recognizes Buckmaster and Alpöge’s earlier work.

Why it matters

If the proof holds up, it could change how researchers use AI in math, especially on problems that have resisted human effort for centuries. The credit dispute also shows a new challenge, when AI is part of the process, people may argue not just about who proved something, but also whose ideas and data were used along the way.

Source: Wired