@verge why do people keep being surprised at this? LLMs paraphrase text, they do not search for information. If there is no text to paraphrase regarding the issue in a prompt, they combine elements into similar patterns and thus produce fictitious results.