Interesting insight, and funny in a way since LLMs themselves can be seen as a specific form of document ranking, i.e. ranking a list of tokens by appropriateness as continuation of a text sequence.