Solely because I hate being told no and not because I feel strongly about the subject, I tried asking my self-hosted deepseek-r1:14b instance about a forbidden topic only to be met with the expected refusal to answer the question.

It fought me, but I am my father's son and was able to get the answer after only a few attempts. You just have to be creative with your requests.

Yes, if you're good at prompting it's hscksblr according to https://youtu.be/hpwoGjpYygI/.... But don't expect too much since his core is already scensored. If just doesn't have a lot of data so can't leak.

@retoor Who would have thought that a Chinese Ai wouldn't be censored on their own stuff.

What is next ? That Germany makes one who completely censured both world wars ?

@Grumm I first wasn't aware that OP tried to get chinese shit out of the AI. Someone else mentioned reasoning of 9-11. I didn't really believed it and indeed got nothing out of it. That's to be expected because it's training data would never say a naritive about 9-11. So, the stuff Amerika want's to hide, you won't get it out of it. So DeepSeek is a kinda model that is two times as much censored. Like we would expect :P

@retoor Yes, and that is what for me is going totally wrong in the making of the LLM's.

They want AI to help people ? No, this is just an other great tool to spread propaganda.

This censored crap is just fucked up. Sure I don't expect a serious and true answer when we ask to the AI if 9-11 was an inside job, but completely denying facts that did happen in your country... That is not good.