Reproduce “RLTLDR: Self-improvement by Internalizing Self-generated Feedback” 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
Michael Kirchhof, Eleonora Gualdoni, Andrew Szot, Khashayar Gatmiry, Aryo Lotfi, Abbas Kazerouni, Omar Attia, Sanjoy Chowdhury and 1 more
When every sampled attempt fails, ordinary reinforcement learning lacks successful examples. RLTL;DR uses verifier feedback, including failed tests, to generate short task-specific insights and trains the policy to predict them from task context. On hard coding and tool-calling tasks with zero successes in 128 tries, it reaches 12–13% Pass@1, versus 0–1% for GRPO, without insights or retries at evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.