If a search engine is full of documents curated directly by the user, through commands to save and index such and such site and this and that document, then a l

If a search engine is full of documents curated directly by the user, through commands to save and index such and such site and this and that document, then a language model is well capable of making judgments on the user's behalf based on their interests. Further, conversations about such documents over time may give priming that narrows the judgments down to what the user approves of the AI's judgments and what they don't. Just speaking from my experience in thinking and working on this concept over the last 2 years...I could be wrong, but testing shows some hope in this idea being a reality. The document set is the critical part here. The user must obtain the documents that are of interest to them. Keeping the document set small and focused allows for higher quality results and discussions about the content.