OpenAI says 10,000 AI agents solved part of a 90-year-old mathematical problem in just 88 hours, although the work has yet to be independently verified.
OpenAI says a new artificial intelligence model and thousands of AI bots have solved part of a notoriously difficult maths problem that has remained unresolved for around 90 years.
The ChatGPT-maker said roughly 10,000 AI agents worked on the Navier-Stokes equations for 88 hours before producing a solution to parts of the problem.
The equations describe how fluids move and are central to understanding phenomena such as turbulence, which remains poorly understood.
OpenAI described the result as a “milestone” and evidence that AI systems are rapidly improving at advanced mathematics. However, its work has not been independently verified or accepted by the Clay Mathematics Institute, which awards the $1 million Millennium Prize for solving a set of major mathematical problems.
The claim has also prompted controversy.
Tristan Buckmaster, a mathematics professor at New York University, said he and Levent Alpöge, a mathematician at rival AI company Anthropic, had been working on the same problem using OpenAI’s Codex tool.
Buckmaster claimed information about their progress had been passed to OpenAI before the company announced its result.
He said OpenAI did not begin working on Navier-Stokes until after information about his and Alpöge’s research had reached the company.
OpenAI denied seeing their work before it was publicly released, saying no user data had been accessed during its research.
Adding that the proofs produced by the two groups differed significantly, the company said: “While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models.”
OpenAI said it began training its new internal model at the end of August and quickly found it was particularly capable at mathematics.
After hearing rumours that two Millennium Prize problems had been solved, researchers decided to deploy thousands of agents to investigate some of the remaining challenges.
By September 5, the agents had produced a result addressing the Navier-Stokes existence and smoothness problem.
The effort involved nearly three million messages between AI agents and 130 billion output tokens.
OpenAI estimated that reproducing the work using its most advanced models would cost about $10 million.
The result resolves two of the four statements required for the full Millennium Prize proof, according to OpenAI.
The company has stopped short of claiming the prize, saying its aim is instead to demonstrate the progress of its AI systems.
OpenAI said: “Our goal in releasing this result is to report on the substantial progress of our AI models. We do not intend to claim the Millennium Prize for this result.”
OpenAI says bots solved 90-year-old maths conundrum in just 88 hours
