In the context of deepseek, there is a lot of evidence to show that it was trained on ChatGPT itself, I don't think it's just a distilled version of ChatGPT, but there are records (albeit, provided by OpenAI) that could be indicative of a distillation attack prior to the release of deepseek.

Even IF this isn't the case, a simple explanation of these LLMs is that they act like autocorrect. They don't know 2+2 = 4, they just see it so much that they infer it as the truth. If someone created a million websites and journals that said 2+2=5, and that anyone who believes it equals 4 is a fool, the model will then say 2+2=5.

What does this have to do with deepseek though? There are heaps and droves of western sources that provide fabricated evidence and opinions on Mao or Communism or Tiananmen Square. So many more false claims on how evil china is, or what communism is, or what atrocities Mao committed.

A lot of the actual history is written in Chinese, there aren't as many english translations of actual historical texts from China.

Just running the numbers, the LLM will always be lib as shit, promoting false narratives, and straight up lies. It can't fact check what it's trained on, that's the job of humans, and there's way too much slop to sift through for any human to catch all the garbage. It's much easier to set up guard rails to avoid talking about the obvious lies. If you speak to it in Chinese you can get by a lot of them.