When Verification Hurts: Asymmetric Effects of Multi-Agent Feedback in Logic Proof Tutoring

1. 1Introduce a knowledge-graph-grounded benchmark for evaluating tutoring systems
2. 2Annotate proof states with step-level feedback and difficulty metrics
3. 3Evaluate the effects of multi-agent feedback on learner performance
4. 4Consider the asymmetric effects of feedback on learner outcomes

AI engineers and educators on a team can benefit from understanding the limitations of large language models in automated tutoring, and how to design more effective feedback systems

💡 Multi-agent feedback in logic proof tutoring can have asymmetric effects on learner performance, and careful design of feedback systems is necessary

💡 Asymmetric effects of multi-agent feedback in logic proof tutoring can hurt learner performance #AI #education