Reproduce “Rubric Curriculum RL: Exploiting the Generation-Verification Gap in Non-Verifiable Domains” 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

Sumeet Ramesh Motwani, Tejas Krishnan, Alistair Letcher, Charles London, Suhaas Bhat, Huitian Jiao, Philip Torr, Riashat Islam and 5 more

Rubric Curriculum RL trains language models for open-ended tasks using relative feedback. A frozen copy of the initial policy compares candidate responses under rubric criteria, while training shifts emphasis to new criteria over time. On FictionBench, RcRL reaches a 73.9% judged win rate against its base model; human preference was 70.5%, with moderate annotator agreement.

Summary by alphaXiv.

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