Reproduce “Rewarding How Models Think Pedagogically: Integrating Pedagogical Reasoning and Thinking Rewards for LLMs in Education” 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
Unggi Lee, Jiyeong Bae, Jaehyeon Park, Haeun Park, Taejun Park, Younghoon Jeon, Sungmin Cho, Junbo Koh and 2 more
PedagogicalRL-Thinking trains a math tutor to plan teaching responses as well as produce them, using Polya’s four-step method and an LLM-based Thinking Reward for pedagogical reasoning. On simulated BigMath tutoring with 7–8B models, the best condition reached a 0.294 Delta Solve Rate, 0.172 Leak Rate, and 0.776 Helpful Rate. Benefits depended on pedagogical prompting and remain unvalidated with human learners.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.