Reproduce “Principled Thoughts for Latent Recursive LLM Systems” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Fahd Seddik, Fatemeh Fard

REST trains latent reasoning handoffs with supervision on transferred thoughts while retaining final-answer cross-entropy. Its losses encourage thoughts to preserve consumer-relevant information, discard irrelevant input, distinguish different outputs, and represent producer uncertainty. In scaled multi-agent experiments, causality plus minimality improved accuracy by 7.5 percentage points over cross-entropy-only at recursion round 3; this configuration does not establish a universal gain.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.