OpenAI recently made waves in the mathematics community by announcing that its AI model had cracked a longstanding math problem that had baffled experts for nearly a century, completing the feat in a mere 88 hours.
However, a professor from NYU has stepped forward, claiming that OpenAI pressured him into assisting with the problem, suggesting that the company might have taken inspiration from his earlier work.
On Tuesday, OpenAI announced it had solved one of the Millennium Prize Problems—seven notorious unsolved math problems identified by the Clay Mathematics Institute in 2000, with a $1 million prize for each correct solution.
The focus of the dispute is the solution to the Navier-Stokes existence and smoothness problem, which investigates whether smooth, three-dimensional fluid flows can remain smooth or develop infinite velocity singularities, according to OpenAI.
Professor Tristan Buckmaster, a researcher from Australia with training in Germany, has alleged that OpenAI may have used his findings after approaching him to collaborate—one of their scientists purportedly hinted that declining the offer could “ruin his career.”
Buckmaster was already collaborating with Levent Alpöge, a promising mathematician based in Long Island working with OpenAI’s rival Anthropic, on a different complex fluid dynamics equation called the Euler problem. During this time, he found a solution using AI resources from both companies.
He revealed that OpenAI representatives reached out to him, seeking a joint announcement for their independent solution to the Navier-Stokes problem.
In a blog post, Buckmaster expressed skepticism about the timing of their announcement, noting that it came shortly after OpenAI became aware of his own progress. It seemed unlikely to him that such a complex problem could be solved “in just a few days.”
Inquiries to OpenAI about whether their AI systems had been exposed to his research were met with silence, which only fueled his doubts over how they could arrive at the solution so swiftly.
Buckmaster noted, “The puzzle is not something one gets to in a few days.” He pointed out that one of the OpenAI scientists had ominously questioned him, “Why would you want to ruin your career?” and suggested a more amicable approach could be possible.
OpenAI’s website featured a post where they attributed the solution to their new AI bot, Astra, claiming it proved that the dynamics of the Navier-Stokes equations can develop singularities in a finite timeframe, thereby solving the problem.
In that post, OpenAI emphasized that no user data had been incorporated in developing the solution but acknowledged they “cannot rule out” the potential use of anonymized data to enhance their models.
“We congratulate Levent Alpöge and Tristan Buckmaster on their remarkable mathematical work,” they stated.
Buckmaster and others questioned the ethical implications of utilizing a customer’s data for competitive advantage.
Bubeck later clarified on a social media platform that he meant to caution Buckmaster against making “unfounded claims” that could damage his career but recognized that his language could have been better.
Interestingly, only one of the Millennium Prize Problems has been solved since the list was created in 1994, when Russian mathematician Grigori Perelman resolved the Poincaré conjecture between 2002 and 2003. He famously declined the accompanying $1 million prize.
In 2018, renowned mathematician Sir Michael Atiyah proposed a solution to another problem, the Riemann hypothesis, but his work was dismissed as it contained critical logical fallacies.
OpenAI has not yet responded to requests for clarification on this matter.
