Uncovering Safety Risks of Large Language Models through Concept Activation Vector

Xu, Zhihao, Huang, Ruixuan, Chen, Changyu · arXiv · 2024