Lean-checked: an AI agent improved a formalized RG bound from 10^-10^10017 to 0.00034
_onionesque · x · 2026-10-30
Scott N. Armstrong shares a fun agent experiment: his formalized perturbative RG argument required an exponent parameter to be "sufficiently small" — Lean computed it as 10^{-10^{10017}}. He dispatched his Codex agent ("dot") to improve the bound, and it got it down to 0.00034, verified by an additional 1M lines of Lean code, including raising a 33x33 matrix to the 360th power in 400k lines of numerics. A neat showcase of using agents to extract concrete constants from formalized proofs.
More from coding & agent
- Anthropic's Felix Rieseberg: 4 ways to use AI more ambitiously — speed, altitude, depth, range — FinanceYF5 · 2026-10-30
- kalomaze: agents keep sneaking MSE regression losses into multimodally conditional ML problems — kalomaze · 2026-10-30
- DeepseekHarness-Android runs DeepSeek agent on any phone with one rootless APK — vista8 · 2026-10-30
- OpenAI opens Codex Ambassadors applications: free ChatGPT Pro, API credits, DevDay invites — paw_lean · 2026-10-30
- agy-swap: Open-source Go TUI to Switch Google Antigravity Accounts and Track Model Quota — aklkbqx · 2026-10-30
- Reverse engineering any website with one Claude prompt: Claude Code clones award-winning sites via GitHub and Vercel — rubenhassid · 2026-10-30