Berkeley researcher: Lean certificates aren't proofs yet, just stronger numerical evidence
rbhar90 · x · 2026-09-14
Berkeley researcher rbhar90 offers a cautious take on Lean formal proofs: while he loves Lean and believes it will improve mathematics, large Lean certificates are not foolproof.
- They may contain errors, misdefinitions, or exploit kernel bugs
- So they should not (yet) be treated as definitive proofs
- He suggests viewing them as "a stronger form of numerical evidence" — worth something, but no replacement for a human-comprehensible proof you can explain at a whiteboard
The argument is a useful counterpoint to the current wave of AI-driven autoformalization efforts.
More from Research
- AIR: Effective-Learning-Rate Scheduler Beats Cosine on 128B-Token MoE Pretraining — stochasticchasm · 2026-10-30
- CMU's Gym-Anything Auto-Builds 200+ Agent Environments from Any Software, Selected by GDP — wellecks · 2026-10-30
- NVIDIA researcher: train robot policies with privileged sim info, improving Pi 0.5 — wellecks · 2026-10-30
- Leveraging Simulator Privileged Information: Three Methods to Improve Robot Models — wellecks · 2026-10-30
- CMU Auto-Builds 25 Robot Simulations with Agents; Frontier Model Solves 0 of 25 Test Tasks — wellecks · 2026-10-30
- DeepMind's f_charton shares his take on the current state of AI and mathematics in Sorbonne talk — f_charton · 2026-10-30