LLMs are pretty decent at 'search' given the inherent knowledge compression, and some amount of inaccuracy is fine.