Any news on what's going on with ernie bot/large language models/AI chat in baidu and elsewhere? It's pretty hard to train a large language model on carefully vetted, correctly censored data, because it needs massive amounts and a diverse variety of language as inputs so it can observe long lists of trends and infer/label the concepts behind each phrase and correlate those concepts with other phrases. But it's not hard at all for a large language model to distinguish between what concepts are and aren't political or in any other of the vague categories that are often censored, because the censorship apparatus has already placed billions of censor:yes/no labels over the last couple decades. So the LLM can check, with relatively strong confidence, whether what it's about to say was frequently censored or not. Those inputs shouldn't be hard for even a mediocre AI engineer to notice that they can implement. And anyone who really wants to dig into the AI to find censored information already has tons of easier, better options for finding censored content.