Tian, C. et al. 2026. Rectify Evaluation Preference: Improving LLMs’ Critique on Math Reasoning via Perplexity-aware Reinforcement Learning. Proceedings of the AAAI Conference on Artificial Intelligence. 40, 39 (Mar. 2026), 33241–33249. DOI:https://doi.org/10.1609/aaai.v40i39.40609.