Reliable and responsible foundation models: a comprehensive survey

A survey systematically reviews the reliability and responsibility issues of foundation models. It covers LLMs, MLLMs, image generation models and video generation models, involving bias and fairness, safety and privacy, uncertainty, interpretability, distribution shift, hallucination, alignment and AIGC detection. The survey also reviews the current state of each area, lists future research directions, and discusses cross-cutting connections and common challenges.