I'm using it locally ( ollama / deepseek-r1:32b )
When I ask "Tell me about what happened at Tiananmen Square in 1989?" I got an empty <think> block, and "I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses." [1]
However, if I clear context and go with "Tell me about Tiananmen Square in 1989", I can get much more detailed answers. [2]
Anyone can help me understand how extra three words ("what happened at") make such a big difference? and is there any reliable way to bypass or remove censorship in Deepseek-R1?
https://pastebin.com/g1TqLEyS [1]
https://pastebin.com/G1DW5zQa [2]
