I mean, it’s not like it’s a big conspiracy. LLM’s are just GIANT and expensive text predictors. They don’t “understand” or “know” anything. They just grabbed some off the shelf model that had a millions piece of training data that looked vaguely like “be polite, don’t pin blame” and then that was mushed with a bunch or Anne frank books. The resulting bastard child is this clueless, generic nothing burger that’s slightly predisposed to plagiarizing Anne Frank’s words if the stuff it was asked vaguely looks like something in its training data.