Steering the CensorShip: Uncovering Representation Vectors for LLM "Thought" Control

Open in new window