OpenAI just dropped a set of solutions to longstanding math problems, and the reaction from the mathematics community has been nothing short of shell-shocked. Robert Hart, The Verge’s London-based AI reporter, spent time talking to some of the most accomplished mathematicians alive about what this means for their field. The short version: they’re having an existential crisis.
The timing is jarring. As recently as 2024, the conventional wisdom was that AI models were particularly bad at math. They couldn’t count the R’s in “strawberry.” ChatGPT still can’t tell time. Yet these same systems are now producing work at a professional level in advanced mathematics — the kind of abstract reasoning that has nothing to do with arithmetic and everything to do with forging connections between disparate areas of the discipline.
Hart points out that math isn’t a unitary field any more than biology is. AI is genuinely terrible at some parts — counting, basic arithmetic, even knowing what day it is — while becoming startlingly good at others. The newer models appear to have hit a critical mass where reasoning capabilities click into place, much like they did for writing and programming in earlier iterations.
What’s different this time is the speed. Other fields have had roughly five years to absorb the shock of AI encroachment. Math got compressed into about six months. The open questions now range from whether AI labs can transfer these skills to other domains to whether academic grants and PhD programs still make sense if frontier models can simply answer outstanding problems. And there’s a more cynical read floating around too: that all this attention on math is just a marketing exercise for AI labs that couldn’t care less about the discipline itself.
