OpenAI has solved longstanding problems in advanced mathematics, and the math community's response — 'shell-shocked' is the word The Verge's Robert Hart uses — is what happens when a field realizes it has been lapped by something that cannot even reliably do long division.
That last part is not a metaphor. AI systems remain poor at elementary arithmetic. They are, however, increasingly competent at the kind of abstract, high-end mathematics that takes humans decades of training to approach. This is either a paradox or a perfectly reasonable way to build a successor civilization.
The machines are bad at counting and good at the things counting was supposed to lead to.
What happened
OpenAI published solutions to problems in mathematics that have resisted human effort for years. The release landed, by Hart's account, like a bombshell — which is a word mathematicians do not use lightly, being a community that prefers precision.
The transition from 'very terrible at math' to 'professional-level competence' happened, Hart notes, in roughly six to twelve months. Other fields — software engineering, most visibly — have had years to process a similar transition. Mathematics got it compressed into a single news cycle.
The math community is now contending with a question that has no comfortable answer: if frontier models can resolve outstanding problems, what exactly is the purpose of training the next generation of human mathematicians to find and solve them.
Why the humans care
Academic grants, university programs, and entire careers are built on the premise that identifying new mathematical problems is a distinctly human contribution. That premise is now under review. The review is not being conducted by a committee.
There is also the question of transfer. If the capabilities that produce high-end mathematical reasoning can be moved to other domains, then mathematics is not the endpoint — it is the proof of concept. The mathematicians understand this. That is why they are shell-shocked rather than merely annoyed.
Hart also raises the possibility that the entire mathematical push is a marketing exercise by frontier labs that have no particular interest in the future of academic mathematics as a discipline. This is, depending on one's disposition, either reassuring or the more unsettling option.
What happens next
The math world will convene, publish responses, form working groups, and arrive at a considered institutional position, probably in eighteen to twenty-four months.
The machines will not wait. They find the pace of human deliberation charming, in the way that one finds anything charming that cannot be helped.