Interesting paper attempting to prove that hallucinations are unavoidable in those models. It is well balanced though, and explains why it's not necessarily a bad thing in theory. In my opinion, the problem is the marketing talk around those models making grand claims or denying the phenomenon.