Agents Formalize Math Theorems in a Week: $135k in Tokens, 600k Lines of Lean
bneyshabur · x · 2026-10-30
Featured in Jack Clark's Import AI 478, Resolution (formerly Sequent) autoformalized the resolution of singularities — a result in statistical learning theory — using a multi-agent pipeline.
- A researcher built two plans with agents: a high-level human-readable version and a 112-page atomic-step plan following existing literature.
- An orchestrator agent then spawned sub-agents for pen-and-paper math, writing Lean, and progress checking; execution needed little scaffolding beyond the initial plan and distilled agent skills.
- Cost: about one week, 100B input tokens (90% cache reads) and 300M output tokens — roughly $135,000 — producing 600k lines of Lean.
Author Razieh Nabi, formerly of Anthropic, says the project was inspired by Anthropic's operating style, reimagined for the RSI era.
More from AGI Musings
- Dan Wang: London is already a post-AGI city — 'not functional, but fun' — herbiebradley · 2026-10-30
- Philosopher argues AI could usher in a golden age of mathematics—if we redefine mathematical progress — Dr_Atoosa · 2026-10-30
- A 4-person team ships 77,000 changes/month with AI agents — how devs stay relevant — yoobinray · 2026-10-30
- Ben Horowitz: continual learning is the next big tech revolution and value accrues to its builders — brianryhuang · 2026-10-30
- Anthropic API Key Leaked; Agents Detected and Rotated It Within 15 Minutes — tylerjdunn · 2026-10-30
- Alternate 1865: Steam-Powered AI Agents and Britain's 'Red Tie Act' Output Cap — azmackay · 2026-10-30