Formalizing CFSG at 3x RH scale: Lean is now the bottleneck for AI math
Vjeux · x · 2026-11-23
Vasily Ilin (Axiom Math) responds to Antoine Chambert-Loir's messy formalization effort, arguing the cleanup itself is valuable — everyone involved now has firsthand experience improving Lean code at scale.
Key points:
- Ilin has argued for months that Lean is the primary bottleneck in AI formalization
- Axiom Math is formalizing CFSG (the Classification of Finite Simple Groups), roughly 3x larger than the Riemann Hypothesis formalization
- At this scale, a full build takes 24 hours; CI breaks, GitHub breaks, Lean tools break or hang, and the PR workflow breaks down
- Conclusion: Lean itself and the surrounding tooling are not prepared for formalizations of this magnitude
The thread is a rare ground-level report on where formal proof engineering hits its limits when applied to mathematics at maximum scale.
More from coding & agent
- 'Permission is the enemy of autonomy': devs slam Codex approval loops over worktree cleanup — Baconbrix · 2026-11-23
- AI agents don't need secrets, they need identities: an SVID-based auth architecture — bibryam · 2026-11-23
- Apple cuts per-dev daily token budget from $300 to $100; cheaper-model strategies follow — Zachly · 2026-11-23
- Fine-tuned Gemma 4 E4B drives 100 villagers offline in a rumor-spread simulation game — AlternativeEnough352 · 2026-11-23
- A lightweight script makes CLI agents debate each other to a consensus write-up — reddit_kwr · 2026-11-23
- Nous ships official Hermes Blender plugin: agents drive live 3D scenes via MCP — Teknium · 2026-11-23