Mathematics has changed. Let's act like it
terrytao.wordpress.com ·
Frank Calegari argues that OpenAI’s October 6 release of more than 700 research papers demonstrates AI’s capacity to make major contributions across mathematics, while also exposing problems with writing quality, attribution practices and correctness in some outputs. He says solved conjectures do not exhaust a field, comparing the moment to the classification of finite simple groups: a landmark achievement that redirected attention but left rich questions and revision work. Hugo Duminil-Copin said the release claimed proofs of every major open problem he had publicly cited.
Calegari contends that mathematics should shift recognition away from merely being first to prove results that machines can generate and toward choosing important problems, developing broad research programs, explaining significance and creating new questions. He recommends evaluating work as potentially AI-assisted, conducting more in-person interviews for graduate admissions and hiring, demanding clearer and more substantive journal publications, broadening graduate education while retaining technical training, and making seminars more focused on context and open directions. He also supports recording autonomously generated, ideally Lean-verified, proofs while expecting authors to explain their mathematical importance. Calegari argues that mathematicians should treat AI-generated results as built on centuries of collective human knowledge and use AI to explore connections and research plans that were previously impractical.
What people are saying
The thread focused less on the article’s slogan than on whether AI-assisted mathematics still counts as mathematics for humans.
What readers liked
- AI-verified proofs can still anchor substantial human scholarship when researchers understand them deeply and use them to open worthwhile new lines…
- Recent AI results on long-stalled questions make the claim that models only harvest trivial leftovers look increasingly detached from the evidence.
Concerns raised
- Proofs that arrive as enormous, ugly, weakly contextualized documents risk becoming mathematically sterile: formally successful but too opaque for a…
- AI may wipe out graduate-level problem niches faster than universities can invent credible new training paths, turning a capability gain into a…
Read full discussion analysis
Source: Hacker News · Based on 15 sampled comments · Analyzed
Discussion overview
The thread focused less on the article’s slogan than on whether AI-assisted mathematics still counts as mathematics for humans. Many commenters accepted that a researcher who selects a meaningful question, fully verifies an AI-produced proof, and turns it into fertile follow-up work has made a real contribution. They treated comprehension, exposition, and steering research programs as valuable work rather than ceremonial leftovers. Several also argued that recent systems have already cracked problems researchers had pursued for years, making dismissals of them as mere symbolic autocomplete look increasingly untenable.
The optimism was tempered by a sharp worry that theorem production is not the same as intellectual progress. Readers pointed to machine-generated proofs described as bloated, opaque, and detached from prior literature, arguing that a result nobody can absorb does little to enlarge the discipline. The most anxious comments centered on careers: graduate students and problem-focused researchers may lose their comparative advantage before academia has created respected new roles. Some saw an industrial leap that should free mathematicians to dream bigger; others argued that faster answers can erase the slow, exploratory path through which humans form concepts, connections, and taste. A further unresolved question was whether agents will soon choose research directions and invent new frameworks, rather than merely prove statements humans already know to ask.
Points of disagreement
The central dispute is whether AI turns mathematicians into more powerful scientific directors or makes their traditional intellectual role economically obsolete. Optimists argue that validated results, new questions, and broad explanation remain substantial human contributions, and that systems already solve problems long beyond neglected routine work. Skeptics reply that opaque proof dumps and automated research planning could remove both thesis topics and the human exploratory process that gives mathematical advances their meaning.
What readers liked
- AI-verified proofs can still anchor substantial human scholarship when researchers understand them deeply and use them to open worthwhile new lines of inquiry.
- Recent AI results on long-stalled questions make the claim that models only harvest trivial leftovers look increasingly detached from the evidence.
- Treating AI as mathematics’ industrial machinery could let researchers explore far more terrain, provided they learn to direct the tool rather than defend old workflows.
- Automated proof search may shift prestige toward explanation, synthesis, and identifying high-leverage questions, making mathematics more socially legible rather than less rigorous.
Concerns raised
- Proofs that arrive as enormous, ugly, weakly contextualized documents risk becoming mathematically sterile: formally successful but too opaque for a community to build on.
- AI may wipe out graduate-level problem niches faster than universities can invent credible new training paths, turning a capability gain into a career shock.
- Faster theorem production can sacrifice the slow route through failed ideas and unexpected connections, which some readers see as where new concepts actually emerge.
- Claims that AI will outperform humans on every mathematical task outrun the evidence, especially for paradigm shifts, novel conjectures, and choosing what is worth pursuing.
Keep reading
Sources and details
terrytao.wordpress.com · Published
Mathematics has changed. Let's act like it
Read report →
Community analysis available for 1 of 1 linked discussions.
Additional discussion searches are still incomplete.
Coverage details and dates
Latest available community analysis:
Summaries: 1 current · 0 awaiting update · 0 not verified
Latest summary update:
Hacker News · terrytao.wordpress.com