"Hallucinations really are a elementary limitation of just how that these designs work nowadays," Turley reported. LLMs just predict the following phrase inside of a response, again and again, "which suggests they return things that are more likely to be legitimate, which is not const