作者位次 #4/10ACL 2026Poster· Natural Language Processing
Curing Miracle Steps in LLM Mathematical Reasoning with Rubric Rewards
Youliang Yuan · Qiuyang Mang · Jingbang Chen · Hong Wan · Xiaoyuan Liu · Junjielong Xu · Jen-Tse Huang · Wenxuan Wang · Wenxiang Jiao · Pinjia He
摘要
In this paper, we observe that current models are susceptible to reward hacking, leading to a substantial overestimation of a model’s reasoning ability. This is evidenced by a high incidence of “false positives”—solutions that reach the correct answer through an unsound process.Through a systemat...