Yoav Goldberg: AI finding long Lean proofs is like md5 collision search — cool but useless
yoavgo · x · 2026-11-22
NLP researcher Yoav Goldberg's cold take on the AI-math trend: machines finding long Lean proofs for hard problems is like machines finding md5 hash collisions — unsurprising (we knew it was hard and doable) and largely useless. He concedes one interesting wrinkle: a human was able to intervene and guide the search, which is much harder to do for md5 collision search — but at its core it remains heuristic search with a very strong heuristic.
More from AGI Musings
- Yudkowsky endorses claim he never said p(doom)=99%, but his 2022 words said "probability roughly 1" — teortaxesTex · 2026-11-22
- Charles Hoskinson: an AI proof of the Riemann Hypothesis should be a century-defining milestone — sudoraohacker · 2026-11-22
- Schmidhuber interview: the universe aspires to ever more complexity — SchmidhuberAI · 2026-11-22
- Simulation startups heat up: 'an extraordinary amount of compute' is coming — pzakin · 2026-11-22
- The evaluation paradox: how do you prove AI improves where it's hard to evaluate? — davidmanheim · 2026-11-22
- Yoav Goldberg: AI finding long Lean proofs is like MD5 collisions—cool but useless — yoavgo · 2026-11-22