If you haven’t already, I strongly encourage you to take a closer look at NotebookLM, a new(ish) Google tool to explore documents via a chat interface, based on a Large Language Model.
It’s worth noting that NotebookLM does allow you to “chat” with documents, it does not train the LLM on those documents, though, so it all lives in the LLMs context window.
I tested NotebookLM a bit over the last few days and must say it struck me as really quite capable and useful.
Among other things, once I let it ingest my whole website including 20+ years of blog archives and found it answered questions well and without any hallucinations I noticed. Another time I threw the full-text PDF of the European AI Act into the system and while it couldn’t tell me the number of pages or words, it did answer questions very well, including complex ones like:
What risks categories for AI systems are specified in the AI Act and who determines the risk for any given system?
Are the EU values laid out in the document?
Does the document lay out the desired dynamics and interplay between government, industry and civil society?
What seem to be areas that the AI Act does not touch upon?
These are decidedly not trivial questions, and as far as I can tell, the answers were spot on.
So as a working tool, this might be a keeper.
