Idea: humans write high-level proofs, AI fills in lemma details in Lean
theshawwn · x · 2026-11-23
Researcher theshawwn proposes a research direction: why are proofs necessarily long? Is there a "macroexpansion" aspect of proving that could be compressed? Instead of tediously proving a list of theorems one by one, accept them as premises, string them together, and let AI fill in each lemma's proof — with the resulting Lean formalization left as an appendix for the curious.
More from coding & agent
- fast-anything open-sources AI-assisted Rust hot-path rewriting: 116x speedup by porting only the slowest 5% — udmrzn · 2026-11-23
- Dev calls out Codex bugs mid-chat with OpenAI engineer: "What have you shipped, Theo?" — tekbog · 2026-11-23
- One-session Opus 5.5 build: Rainline, a procedural neon driving game in a single HTML file — heypearlai · 2026-11-23
- Stop asking Gemini nicely: a post-answer checkpoint hook kills boilerplate follow-ups — bfrench · 2026-11-23
- Free-running personal agents can nerdsnipe you at any moment, researcher warns — MikePFrank · 2026-11-23
- DTC brand builds 3D unboxing simulator with Perplexity, live packaging costs — jeffgrimes9 · 2026-11-23