LLMs are not smart. LLMs do not know anything. They are token predictors. They are autocomplete with extra steps. Nothing about what comes from them is tied to the real world, just what text has appeared with what other text most often within a context window.