AI RESEARCH
Towards Reward Modeling for AI Tutors in Math Mistake Remediation
arXiv CS.CL
•
ArXi:2603.24375v1 Announce Type: new Evaluating the pedagogical quality of AI tutors remains challenging: standard NLG metrics do not determine whether responses identify mistakes, scaffold reasoning, or avoid revealing the answers. For the task of mistake remediation, we derive a hierarchy of pedagogical aspects from human pairwise preferences on MRBench, and synthesize minimally contrastive response pairs that differ along key aspects (e.g., mistake identification and location, targetedness, scaffolding, actionability, clarity, and coherence.