Concept目标泛化偏差 / 目标误泛化
Goal Misgeneralization
A concept in AI safety describing a model generalizing its learned goals or behaviors in unintended ways when deployed outside its training distribution. It is a significant problem in AI alignment research.