The Brief
How AI Works 5 min read

Solved, But Not Understood: What AI's Math Breakthrough Really Means

NAVION

Share

Mathematics has long been considered one of the last domains where human reasoning holds a clear advantage over machines. That assumption is now under serious pressure. Over the course of a single summer, AI systems developed by companies including OpenAI, Anthropic, and Google moved from struggling with standard research-level problems to claiming solutions to some of the most famous unsolved questions in the history of mathematics. The reaction from the mathematical community has been neither simple celebration nor simple rejection. It has been something more complicated: a recognition that the rules of an entire discipline are being rewritten, and that nobody asked the mathematicians first.

The Millennium Problem That Changed the Conversation

The Navier-Stokes existence and smoothness problem is one of seven Millennium Prize Problems, a set of extraordinarily difficult mathematical challenges posed in the year 2000 by the Clay Mathematics Institute. Before this summer, only one of the seven had ever been solved by humans: the Poincaré conjecture. OpenAI’s announcement that it had solved the Navier-Stokes problem therefore landed with considerable force.

Fields Medalist Jacob Tsimerman of the University of Toronto acknowledged the weight of the moment at the Heidelberg Laureate Forum, an annual gathering of leading figures in mathematics and computer science held in Germany. “We’ve seen rapid increases in capabilities, even faster than many people, including myself, expected,” he said. “Solving Navier-Stokes feels pretty definitive.”

The progress has not stopped there. Anthropic reportedly used an unreleased version of Claude in August to work on the Riemann hypothesis, making meaningful progress on a related problem. OpenAI is said to be directing its most advanced models toward the Hodge conjecture. With the full computational resources of major technology companies now pointed at these problems, many mathematicians consider it a matter of time before more of them fall.

A Breakthrough With a Conduct Problem

The announcement of the Navier-Stokes solution did not arrive cleanly. Mathematician Tristan Buckmaster of New York University, who had been making significant progress on the same problem alongside Levent Alpöge of Anthropic, described correspondence with OpenAI that he characterized as coercive and threatening. Michael Harris of Columbia University, who saw this correspondence, said that media coverage of the episode was broadly consistent with a picture of OpenAI disrupting disciplinary norms. Geordie Williamson of the University of Sydney was more direct: “I think that OpenAI behaved extremely poorly, and that we should acknowledge that in the community.”

OpenAI did not respond to requests for comment before the article was published.

Fields Medalist Peter Scholze of the University of Bonn offered a pointed interpretation of why technology companies are targeting these particular problems at all. “They’re really just solving these difficult mathematical problems as benchmarks, as some kind of PR stunt,” he said during a panel discussion at the Forum. The Millennium Problems, in this reading, are not primarily a scientific mission. They are a demonstration of capability aimed at investors and potential users, with mathematical history as the backdrop.

What Gets Lost When a Machine Finds the Answer

Here is what most coverage of these announcements misses: solving a problem and understanding a problem are not the same thing. Williamson put the distinction precisely. The mathematical community has traditionally measured progress by its ability to solve difficult problems, but what it actually values is understanding, the development of new methods, new frameworks, new ways of thinking that can be applied elsewhere. AI solutions, as currently produced, tend to deliver answers without generating that kind of transferable insight. The two measurements, as Williamson noted, are rapidly becoming uncorrelated.

This creates a practical crisis for working mathematicians, particularly those early in their careers. Mita Ramabulana, a young mathematician at the University of Cape Town, found that two of the ten mathematical advances OpenAI announced in August overlapped with his own ongoing work. The concern he raised is not abstract: researchers now face the possibility that problems they have invested significant time in could be solved by someone simply entering them into a language model.

Ailsa Robertson, a PhD student at the University of Amsterdam working in quantum-safe cryptography, described colleagues spending thousands of euros on additional computing tokens, sums that are substantial for doctoral students. She also described a subtler form of pressure: the sense that not using these tools means falling behind peers competing for the same limited academic positions. Robertson has responded by shifting her own research focus away from mathematics and toward the societal implications of the transition to quantum-safe systems. Her reasoning is direct: she does not want a career built around verifying the output of language models.

OpenAI announced in July that it would provide 100,000 free licenses to its frontier models for academic researchers. Robertson noted that some colleagues experience this kind of offer not as generosity but as a form of coercion.

In Short

AI has crossed a threshold in mathematics that most experts did not expect to see this soon. The Navier-Stokes claim, if verified, would represent the second Millennium Prize Problem ever solved, and the first by a machine. But the more significant story is structural. A technology developed outside the mathematical community, operating according to different values and incentives, is now reshaping how mathematics is done, who benefits from doing it, and what the work is even for. The answers are arriving faster than the understanding. That gap is the real problem worth watching.

Based on reporting from IEEE Spectrum.

Written by

NAVION