Yoav Goldberg: AI finding long Lean proofs is like MD5 collisions—cool but useless
yoavgo · x · 2026-11-22
Commenting on the wave of AI solving hard math problems via Lean, prominent NLP researcher Yoav Goldberg offers a skeptical take: machines finding long Lean proofs for problems he doesn't care about is like machines finding MD5 hash collisions—cool that it can be done, but unsurprising (we know it's hard and know it's feasible) and ultimately rather useless.
Related event: Yoav Goldberg: AI's Long Lean Proofs No More Surprising Than Breaking MD5(2 posts)→
More from AGI Musings
- Ex-NSA's Perlroth: AI is multiplying cyber risk, exploits now beat patches by 7 days — chrisrohlf · 2026-11-22
- Szegedy predicts proofs without formal verification won't count within a year — aran_nayebi · 2026-11-22
- Quant firms 'print money' with reasoning models — and early access looks like cheating — scaling01 · 2026-11-22
- Two compute costs that decide AI takeoff: training the improver and paying for its exploration — 1a3orn · 2026-11-22
- Largest Lean proofs cost at most ~$50 in compute, per Claude's estimates — 1a3orn · 2026-11-22
- a16z: LLMs update predictions in mathematically precise ways — but AGI needs continual learning — vishalmisra · 2026-11-22