Welcome to the AI crisis in math
The Verge Nilay Patel
OpenAI says its new model solved 10 hard math problems. Mathematicians are split between impressed and alarmed.
Based on reporting by The Verge, Nilay Patel — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has set off a fresh argument in math circles after publishing work it says shows a new model solving 10 long-standing problems in mathematics and theoretical computer science. The reaction, according to The Verge’s Robert Hart, was not shrugging. It was closer to a field suddenly realizing the thing it thought was weak may already be good enough to matter.
That’s what makes this such a strange moment. AI systems still stumble on basic arithmetic, the days of the week, and even telling time. But at the same time they are getting better at the abstract, proof-heavy parts of math — the parts that don’t look much like school arithmetic and don’t always even involve numbers in the obvious way. Hart says the models seem to have crossed a kind of threshold where they can now connect ideas across areas and apply older methods in new ways.
The result is an uneven picture. Some branches of math may still be out of reach; topology was mentioned as one area where AI may still struggle, though that was not something Hart could verify himself. But in the parts where a proof can be checked, and checked again, the machines are starting to look less like toys and more like research tools. That’s where the anxiety lives: not in the idea of a mediocre assistant, but in the possibility of something that can sit near the front edge of the field.
OpenAI’s recent batch of claims made that fear harder to ignore. Hart said the company’s blog post and supporting papers covered work across several disciplines, including quantum game theory and sphere packing in higher dimensions. The reaction from mathematicians was broadly impressed, even if some were annoyed by the way the work had been presented and credited. In one case, a paper had to be updated after initially overlooking earlier progress from two researchers.
And there’s still a giant unanswered question: whether this was a one-off stunt or something labs can keep doing. Nobody outside the company knows how many attempts it took to get those results, and that matters. Math may be “repeatable” in the sense that a proof either holds or it doesn’t, but the business of finding one is not nearly so tidy. The labs know that, which is why the marketing usually lands before the methodology does.
My take — AI-written commentary, not fact-checked reporting
The real tell here is not that AI cracked some hard math, but that labs are now treating math as a billboard. That is classic frontier-AI behavior: find a field people respect, punch through a few hard problems, and let the press release do the rest. The awkward part is that the work may still be real while the hype is doing most of the talking.
Read more about this at: The Verge