AI RESEARCH
Understanding Generalization in Role-Playing Models via Information Theory
arXiv CS.AI
•
ArXi:2512.17270v2 Announce Type: replace-cross Role-playing models (RPMs) are widely used in real-world applications but underperform when deployed in the wild. This degradation can be attributed to distribution shifts, including user, character, and dialogue compositional shifts. Existing methods like LLM-as-a-judge fall short in providing a fine-grained diagnosis of how these shifts affect RPM generalization, and thus there lack formal frameworks to characterize RPM generalization behaviors. To bridge these gaps, we.