OpenAI claims its AI solved Navier-Stokes in 88 hours

IA solves Navier-Stokes millennium problem: milestone or plagiarism?

OpenAI claims to have solved one of the Millennium Prize problems. Or so they say. Its AI, coordinating 10,000 agents and operating over 88 hours, allegedly demonstrated the Navier-Stokes equations—the challenge mathematicians have pursued since the 19th century. However, the community is unconvinced: between accusations of plagiarism, multi-million dollar offers, and threats, this announcement could be either the decade's greatest scientific achievement or the biggest intellectual fraud in recent history.

OpenAI's announcement: 88 hours and 10,000 agents

OpenAI asserts that its model reached a solution after 88 hours of work with around 10,000 coordinated AI agents. The company hurried to stress that the demonstration was verified using Lean, a formal mathematical proof assistant, which, theoretically, should protect the proof against logical errors. According to reports, the final demonstration spans 165 pages.

The announcement quickly sparked intense debate. While some hail it as a historic breakthrough, others point out that the AI relies on tools programmed by humans, such as Lean, and its role in the actual discovery is, at best, questionable. OpenAI's refusal to share the initial prompts needed for replication does little to help.

Navier-Stokes: what's at stake

The Navier-Stokes equations describe the movement of fluids: water, air, blood. Solving this Millennium Prize problem means accurately understanding, predicting, and calculating how they flow. It is the basis of meteorology, aeronautics, and medicine. But the path to a solution is fraught with nuances: as noted, the word 'exactness' is a cardinal sin in this context; most practical applications are satisfied with numerical approximations.

The shadow of plagiarism: Spanish mathematicians

The first sign of trouble emerged with suspicions that the original idea belonged not to OpenAI, but to two Spanish mathematicians: Diego Córdoba and Luis Martínez-Zoroa. According to arguments presented, these researchers had spent over a year working on a simplified version of the problem that formed the basis for the solution when scaled up with AI. The procedure used by OpenAI on the major problem was allegedly 'very similar' to their training sessions, leading some to speak of outright crime.

The Buckmaster plot: million-dollar offer and threats

The controversy escalated when it emerged that the AI allegedly plagiarized the work of mathematicians Levent Alpöge and Tristan Buckmaster. According to reports, OpenAI representatives, when confronted, allegedly offered Buckmaster one million dollars in exchange for appearing as a co-author and endorsing the official version, while excluding Alpöge due to his affiliation with a competitor. If he refused, they allegedly threatened to 'end his career.' The episode, reminiscent of a mob film, has brought into focus doubts about industry methods.

Milestone or smoke? Skeptical arguments

The rebuttal soon trinc in the form of a litany of grievances. The lack of reproducibility—OpenAI has refused to publish the prompts—the suspicion that the AI 'fed' upon external data without permission, and the precedent set by the famous P-NP case, where a purported resolution ended in error, fuel caution. Even Terence Tao's presence on the most famous mathematics blog has been interpreted as a sign that the community is taking it seriously.

In any case, some recall Grigori Perelman: the Russian mathematician who solved Poincaré's conjecture and declined the Fields Medal and the million dollars to live with his mother in St. Petersburg. An example that glory and money are not always coupled with rigor.

If the demonstration is confirmed, we will live through a before and after in mathematics. If it collapses, it will be the greatest intellectual fiasco of the digital age. One thing is certain: AI has already burst into the realm of rigor, and not quietly.