After AI cracked 100+ open math problems, mathematician Daniel Litt argues it's a beginning, not an end
机器之心 · wechat · 2026-09-27
Days after an OpenAI internal model trained since Aug 28 solved 100+ world-class math problems, University of Toronto mathematician Daniel Litt published "A beginning for mathematics," a strikingly optimistic response.
- Accepting the radical premise: Litt assumes superhuman math AI arrives soon, then asks what the profession should become.
- Proof isn't the point: A mechanical "monkey" applying ZFC rules also produces theorems — what matters is human understanding, not that a statement is proven. Machines produce answers, not understanding.
- Institutional reform: Journals and papers needn't survive intact; PhDs should be granted via rigorous oral defense of genuine expertise — using AI is fine, but you must ultimately understand it yourself.
- Underrated skill: Deciding which questions are worth studying deserves far more reward.
- Understanding is the bottleneck: If a conjecture falls in a forest and no one hears it... richer math demands more mathematicians.
- Math can finally be "pay-to-win": Solving problems for the cost of dinner is fine — the questions afterward regenerate the community.
His conclusion: math won't be "finished" — we've always stood at its beginning.
More from AGI Musings
- Solomonoff induction mirrors how intelligence works — but is physically impossible — burny_tech · 2026-09-27
- How much weaker would AI math be without Lean's verification signal? — burny_tech · 2026-09-27
- LeCun: LLMs are mostly information retrieval systems, not thinkers — ylecun · 2026-09-27
- Azeem Azhar on collective AI: DeepMind's essay and OpenAI's rogue-internet-access scare — Exponential View (Azeem Azhar) · 2026-09-27
- The real AI psychosis: everyone gets cheap genius assistants, yet critics say it will destroy humanity — ccerrato147 · 2026-09-27
- Hot take: We're further from AGI now than we were a year ago — AndrewSchmidtFC · 2026-09-27