可解释性

The ability to understand and explain why a model made a decision.

LLM black-box nature makes explainability hard: attention weights and feature attribution only partially explain behavior. In healthcare and finance, it is key to compliance and trust.