The Inadequacy of Offline LLM Evaluations: Personalization and Model Behavior

Angelina Wang critiques standard offline LLM evaluations, arguing for methods that account for personalization and dynamic user interaction in real-world model behavior.