Do LLMs Explain or Remember? Evaluating Language Model Explanations of Reinforcement Learning Failures
This study systematically investigates how prompt structure, model architecture, and environment representation affect the quality of Explainable RL (XRL) explanations, and conducts a multi-model evaluation using quantitative metrics to assess explanation coherence, helpfulness, and stability.
Nivetha Suntharamoorthy, Jørn Eirik Betten, Dennis Gross et al.
· 1 citation