OpenAI’s Navier–Stokes Claim Sparks a Fight Over Who Got There First

OpenAI’s Navier–Stokes Claim Sparks a Fight Over Who Got There First
On Monday, NYU mathematician Tristan Buckmaster posted work with Levent Alpöge, an Anthropic researcher collaborating in a personal capacity, showing that a simplified Navier–Stokes system could break down. The pair had spent nearly a year using publicly available AI tools on related questions. Navier–Stokes—the equations governing fluid flow—is one of the Clay Mathematics Institute’s seven Millennium Prize Problems, each carrying a $1 million award.

OpenAI says it began its own push on September 1 after hearing rumors of progress on Millennium problems. By Saturday, it says, roughly 10,000 coordinated agents using an unreleased model had generated a full proof in about 88 hours, with another 17 hours of verification. The company’s executives put the compute cost in the millions; the Clay institute had not commented on the proposed solution. Greg Brockman called it “a major milestone for both AI and mathematics,” while Sam Altman described watching the work unfold as one of OpenAI’s most amazing moments.

But Buckmaster said he learned that information about his and Alpöge’s progress had reached OpenAI. Their chosen route, he argued, was unusual: “It is not the direction one arrives at in a few days by giving a model the problem statement.” He stressed that he had not seen OpenAI’s proof and did not know whether the pair’s data had been used. His account also alleged pressure to exclude Alpöge from authorship because of his Anthropic affiliation—an allegation Altman disputed, saying he “never ever asked for Levent to be removed from authorship.”

OpenAI denies that researchers or agents saw the mathematicians’ work before it was public, or accessed specific user data. Yet it acknowledged a crucial qualification: it “cannot rule out” that de-identified data from use of its products helped improve its models.

That hedge has fueled wider unease. Yann LeCun amplified calls for “alternative points of view,” and a retweeted critique congratulated OpenAI for not letting “customer transcripts go unmined.” The issue now reaches beyond one proof: whether academics can safely use proprietary AI systems when the firms behind them can mobilize private models and millions of dollars in compute to pursue the same discoveries.

Continue reading https://foxvector.com/stories/01a089d7-27d1-0563-72d5-225d979881e3