3 ms·
It seems to me that thinking models are harder to decensor, as they are trained to think whether to accept your request.
by shikon7 11mo ago
It seems to me that thinking models are harder to decensor, as they are trained to think whether to accept your request.
- int_19h 11mo agoIt goes both ways. E.g. unmodified thinking Qwen is actually easier to jailbreak to talk about things like Tiananmen by convincing it that it is unethical to refuse to do so.