Running on Zero 2 DeepSeek-R1 Censorship Steering π³ 2 Generate text with adjustable censorship control
Running on Zero 8 Refusal Censorship Steering π¦ 8 Generate text with adjustable censorship control
Steering the CensorShip: Uncovering Representation Vectors for LLM "Thought" Control Paper β’ 2504.17130 β’ Published Apr 23 β’ 1