Imho, that's an argument to not censor the models. Censoring will train users to assume that anything wrong has already been filtered, which means if something

Imho, that's an argument to not censor the models. Censoring will train users to assume that anything wrong has already been filtered, which means if something slips through and they get misinformation they will be more likely to uncritically take it as truth - after all, the machine wouldn't be allowed to answer otherwise. Let it spit all the bullshit (together with the good things) it has been trained on. People will learn far better that media literacy and cross-checking are still needed.