OpenAI’s math breakthroughs send shockwaves through the discipline

Robert Hart, AI writer at The Verge in London, talked with several of the world’s leading mathematicians and came away with a picture of a genuine existential crisis. The immediate trigger was OpenAI’s announcement of solutions to open problems in higher mathematics, a move that exploded in the community like a bomb and ignited a heated debate about the very nature of the profession. Mathematicians now wonder not only what their role will be, but what mathematics is at all when a model can close research gaps that have lasted decades in a short span.

The shift arrived in extreme compression. Up to half a year to a year ago, the consensus was that large models failed at mathematics in an embarrassingly obvious way. The classic illustration was counting the letter “R” in the word “strawberry”, a task that models repeatedly botched. In a very brief interval the systems leapt from a hobbyist level to professional-grade performance on abstract mathematics, a trajectory that other domains such as software engineering have taken over many years. Hart stresses that the leap was not accompanied by a parallel upgrade in fundamentals: the models still fail at basic arithmetic, on days of the week and at reading a clock, a phenomenon documented by Elissa Welle months ago.

That disconnect creates a stark paradox. To be good at mathematics you are supposed to be able to count, add and multiply, yet the new models skip that layer and go straight to complex proofs. Hart notes that even the “R-in-strawberry” problem was probably solved through hard-coded point solutions rather than genuine counting ability. The gap between failure on elementary tasks and success on Olympiad-level problems or research papers upends assumptions about how “mathematical thinking” should operate in a machine.

The practical question that now haunts academia concerns money and manpower. If frontier models solve open questions, what is the purpose of research grants and training programs for a new generation of mathematicians who are supposed to spot problems and solve them? There is a fear that big labs are using mathematics as a marketing exercise—a showcase of capability that looks impressive outwardly, without real interest in the fate of one of the oldest foundational academic fields. Hart heard a range of opinions, but the bottom line remains open: are we witnessing a powerful assistive tool, or the beginning of a process that empties the profession of content?