On Tuesday, OpenAI announced that it had found a solution for a famous unsolved problem in mathematics by using a swarm of advanced AI agents. But the occasion quickly unraveled into an explosive scandal, with mathematicians accusing the company of stealing their work.
OpenAI’s purported solution was for one of the so-called Navier-Stokes equations, which mathematically describe how fluids, ranging from wind to ocean currents, move and behave. These differential equations — equations that describe how different quantities change in relation to others — have been used for nearly two centuries with reliable results, but scientists have long wondered if there are scenarios where their predictions break down and describe something that’s physically impossible. In 2000, the Clay Mathematics Institute listed the Navier-Stokes as one of its six Millennium Problems, with a prize of $1 million for being the first to solve each one.
Enter OpenAI, which claims it found a way to break the Navier-Stokes by unleashing 10,000 of its AI agents on the problem for 88 hours, costing millions of dollars in computing costs, according to OpenAI’s head of research Mark Chen. The long and short of some exceedingly complicated math was that the legions of AI agents identified an apparent scenario in which the equations can describe fluids accelerating to an infinite velocity — resulting a so-called “blowup” or singularity. This, as far as we know, is impossible, thus demonstrating that the equations have a breaking point.
But hours before OpenAI’s formal announcement came an explosive claim from the NYU mathematician Tristan Buckmaster, who had been working on the problem for years before OpenAI got involved, and was close to publishing his own solution.
In a lengthy public statement, Buckmaster accused OpenAI’s solution of following a suspiciously similar path to his own, and said that the AI company only poured its resources into the Navier-Stokes after recently learning about the progress Buckmaster and his colleague Levent Alpöge, a mathematician at OpenAI’s biggest rival Anthropic, had been making.
Adding to the suspicion, Buckmaster said he had been using OpenAI’s Codex tool to help check his work, insinuating that the company accessed his sessions, accidentally or intentionally, to copy it. When Buckmaster asked OpenAI about this, he didn’t get an answer, he claimed.
OpenAI’s response to Buckmaster’s allegations only added fuel to the fire. While denying that its researchers and AI agents accessed or saw the mathematician’s work before it was publicly released, OpenAI also admitted that it “cannot rule out that de-identified data derived from their usage of our products helped improve our models.”
That’s not all. Buckmaster also alleges that OpenAI mathematician Sébastien Bubeck pressured him to cut Alpöge as an author on his paper, which Buckmaster implies is because Alpöge worked for Anthropic. When Buckmaster refused and threatened to go public, Bubeck allegedly replied, “Why would you ruin your career?”
Bubeck has denied Buckmaster’s version of events, calling them “false and inflammatory” in a public statement.
The academic community has largely been appalled by what they’re seeing.
Terence Tao, a renowned mathematician and leading authority on differential equations, excoriated what AI was doing to the field.
“We have now seen that even the rumor of someone working on a problem can trigger a massive amount of AI-powered effort to flatten it before the original research project has time to reach its full potential,” Tao wrote on Mastodon. “The incentives may now be pointing in the direction of no longer sharing any promising research directions with the broader community, which would reverse centuries of traditions of open science and do serious long-term damage to the future of the field.”
More on AI: OpenAI Denies Coverup After Rogue Swarm of Agents Reportedly Targeted a Second Site From Hugging Face
