On October 6, OpenAI released over 700 research papers claiming results across multiple areas of mathematics. The reaction split the profession. Some mathematicians see a catastrophe; Frank Calegari, writing as a guest on Terry Tao's blog, sees something closer to the 1983 classification of finite simple groups — a massive technical achievement that cooled interest in a field precisely because outsiders confused "solved the big problems" with "finished the subject." His argument is that mathematicians are grieving the wrong thing. Calegari's core claim is structural, not sentimental. The OpenAI papers are, by his own account, "atrociously written," violate community attribution norms, and some are simply wrong. But the problems they target — even in the Langlands program — are boundaries of understanding, not the understanding itself. He quotes Hugo Duminil-Copin's lament that every major open problem he'd publicly mentioned was claimed solved in one dump, then reframes it: this is what it felt like to be a finite group theorist in 1970 if someone scattered the classification proof across your office floor on a spirit duplicator. The generative heart of the essay is Calegari's argument that AI lowers the barriers to pursuing speculative, cross-domain mathematical programs that were previously untouchable. He offers a specific example from his own research: a connection between commensurability classes of compact arithmetic hyperbolic three-manifolds, elements in the Bloch group K₃(K), and a hypothetical geometric Kolyvagin system linked by an unknown integral arithmetic Jacquet–Langlands correspondence. He never pursued it because the conditional overhead was too high and assembling the technical background would have consumed years. AI changes that calculation. Calegari is clear-eyed about what's being lost. The mountaineering model of mathematics — where prestige accrues to whoever summits first — is devaluing fast. He quotes Christian Szegedy: "That specific kind of heroism will matter less, and other skills will matter more: choosing which terrain is worth exploring, steering fleets of airplanes, and explaining what they find to the rest of the world." The profession's reward structures — hiring, tenure, publication — were built around the old model. The essay is implicitly a call to rebuild them. The ugly coda is honest. Calegari notes that even as the community debates what mathematics should become, many mathematicians are already racing to crank out AI-assisted papers to claim priority. The gap between producing novel proofs and demonstrating understanding is widening in real time. The piece ends abruptly — possibly truncated — but the implication is clear: the profession's revealed preferences are running ahead of its stated values. What makes this essay matter beyond mathematics is its generalizable insight: when AI solves the legible problems in a field, the temptation is to declare the field dead rather than recognize that the legible problems were never the point. The real work — building structural understanding, connecting distant areas, formulating new questions — was always harder to see and harder to reward. AI doesn't eliminate that work. It makes the failure to do it more visible. Calegari's position is ultimately generous but demanding. He refuses both doomerism and triumphalism. The profession needs revised classification efforts (like Gorenstein-Lyons-Solomon), new reward structures that value breadth and structural ambition over narrow technical firsts, and the intellectual honesty to admit that walking around with your eyes closed at ground level won't help you see anything from the air either.
