Mathematicians say AI proofs are incomprehensible; giffmana proposes a readability reward model
giffmana · x · 2026-09-21
- Mathematicians complain that AI-generated proofs are incomprehensible and don't advance mathematical understanding.
- DeepMind's giffmana calls it the typical "make it work, then make it nice" cycle — coding is further along, math will follow.
- He suggests a tractable fix: train a reward model/judge measuring proof readability and use it for post-training or guidance.
Related event: Mathematicians Complain AI Proves Are Unreadable; Researchers Debate Fixes(4 posts)→
More from AGI Musings
- 'Eigenism is communitarianism dumbed down for nerds': philosopher pushes back on Hendrycks — edelwax · 2026-09-21
- Uncertain about insect consciousness? Author argues we should still stop torturing them on farms — NathanpmYoung · 2026-09-21
- AI enters the Riemann Hypothesis arena: when machines prove what humans can't grasp — ugail · 2026-09-21
- antirez: Jev hype shows the AI bubble can't tell what actually matters — antirez · 2026-09-21
- AI safety researcher rewatching Terminator 2: surprisingly good take on AI extinction risk — JeffLadish · 2026-09-21
- "Understand neuroscience and you can't believe humans are conscious" — a reductionist hot take — burny_tech · 2026-09-21