OpenAI says it cracked 90-year-old maths problem in 88 hours

OpenAI claims it has cracked a 90-year-old mathematics problem in just 88 hours using an advanced AI model and thousands of AI bots. The company focused a group of approximately 10,000 AI agents on solving a notoriously difficult problem known as the Navier-Stokes equations, which pertain to fluid motion. This problem had eluded proof for 90 years, with the underlying argument for a correct mathematical equation still unknown.

OpenAI labeled the solution they discovered as a "milestone" and saw it as evidence of the rapidly improving capabilities of AI tools. However, the solution has not yet been independently verified or publicly accepted by the Clay Mathematics Institute, a U.S.-based math organization that runs the prestigious Millennium Prize. The prize offers a $1 million reward to the first person to solve certain mathematical challenges.

OpenAI started training a new model in late August that showed promise in handling mathematics. Although the new model is only accessible within the company, OpenAI researchers decided to use it on notable advanced math problems. The breakthrough came on September 5, 88 hours after OpenAI assigned 10,000 AI bots to the task. The AI bots exchanged nearly 3 million messages and utilized 130 billion output tokens to solve the Navier-Stokes existence and smoothness problem, which is central to understanding turbulence.

OpenAI claims the effort would have cost approximately $10 million based on its pricing for the output from its most advanced models. While the company did not claim the $1 million Millennium Prize, the solution addresses two of the four statements in the proof that the prize demanded. However, the claim has sparked controversy, as some mathematicians argue OpenAI's work was influenced by information about Buckmaster and Alpöge's own progress on the problem.

OpenAI maintains it was unaware of their work until the solution was made public and that no user data was accessed in their research.