UCLA PhD's LLM agents produce 126K-line Lean 4 proof of MIP* = RE core theorem in 63 days
siyan_zhao · x · 2026-10-06
Yanqiao Zhu, a final-year CS PhD at UCLA, is on the job market for research roles and highlights his LLM agent work:
- RetroAgent (COLM 2026): an LLM agent that searches over structured memory for multi-step retrosynthesis planning, trained with RL from verifiable rewards.
- FormalFlow: agents coordinating via a shared repo, CI and code review produced a complete machine-checked Lean 4 proof of a core theorem underlying MIP = RE in 63 days — about 126K lines — and repaired two errors in the published statement.
- ARLArena (ICML 2026): an agentic RL training platform.
He will attend COLM 2026 in SF, Oct 5–9.
More from coding & agent
- Claude can now join Slack group DMs and answer in threads like a teammate — EricBuess · 2026-10-06
- Google details 2-tier agent memory architecture on AlloyDB, cutting token spend up to 70% — rseroter · 2026-10-06
- Compound rewrites Excel, PowerPoint and Word in Rust entirely with AI, claims better performance than Google Workspace — peterjliu · 2026-10-06
- Thinking Machines open-sources Prompt Distillation recipe to bake long system prompts into model weights — TheZachMueller · 2026-10-06
- Claude can now join Slack group DMs and follow threads like a teammate — maxleiter · 2026-10-06
- opencode-leaner plugin prunes coding agent context from ~10.3k to ~1.3k tokens — laerciosantana · 2026-10-06