Reproduce “LLM-as-a-Tutor: Policy-Aware Prompt Adaptation for Non-Verifiable RL” 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
Yujin Kim, Namgyu Ho, Sangmin Hwang, Joonkee Kim, Yongjin Yang, Sangmin Bae, Seungone Kim, Jaehun Jung and 2 more
LLM-as-a-Tutor adapts open-ended instruction-following reinforcement learning by having a tutor compare two current-policy responses and append a requirement with matching rubric criteria when their quality seems indistinguishable. In tests using a Qwen3-1.7B-Thinking policy and Qwen3-8B-Thinking tutor and judge, it reports five-metric average 51.96 ± 0.226, versus 51.04 for policy-adaptive rubrics and EVA. It does not lead on every component metric.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.