I’ve been using Google’s NotebookLM to help me complete my long-delayed book on Patent Litigation.

NotebookLM is a RAG (retrieval augmented generation) system that Google developed with Steven Johnson, well-known author of books on invention and innovation.

You can upload documents to NotebookLM — or, more recently, provide it links to your docs on Google Drive or on websites — and NotebookLM will use your constrained collection of documents as the basis for new AI-generated documents, and for chats. Each doc will automatically be summarized, as will  the entire collection. You can also generate other materials, such as a study guide, briefing doc, FAQ, and timeline. [Later, describe how different from doing within Google Drive with Google’s Gemini LLM, or from using a project in Anthropic Claude.]

Among the most interesting automatically-generated output are an “Audio Overview” and a “Mind Map”.

From my 36 linked files (“Sources” in NotebookLM), many of which contained very rough bullet-pointed notes to self, it generated a 21-minute audio conversation between two synthetic characters, who might sound like clueless marketing folks, but who have been given a very good script to follow:

https://notebooklm.google.com/notebook/811dfa64-7bb8-430a-ac60-42a62315ea49/audio 

I’m ashamed to say that these two fictional characters do a better job of summarizing key points and nuances of my forthcoming book than I’ve been able to do myself. [List 2 or 3 nuances captured.] The audio generation appears to be greatly improved when from I last tried this in Nov. 2024.

See a transcript of the NotebookLM-generated audio, with my comments and clarifications, and highlighting a few places where I thought NotebookLM did especially well, making some of my key points better than I did. I tried the audio beta “Interactive mode” to correct a small error, but it seemed to get the two speakers unnecessarily sidetracked. (By the way, it’s slightly awkward to get a text transcript of the audio within NotebookLM: you designate the audio file as a new source, and then clicking on it within the left-hand Source panel will reveal the transcript.)

NotebookLM also provided a MindMap of my notes for the Patent Litigation book, part of which is shown below, expanding some of the higher-level locations in the map:

Again, this was generated by NotebookLM with no involvement from me other than providing links to my docs (which of course I had to write, over several years), and pressing the “MindMap” button. As author of the underlying docs, I not only feel this automatically-generated map is accurate, and fairly complete (though I would be shocked if there isn’t something important missing, or something that’s been given too much or too little emphasis), but also useful in thinking how to whip all my docs (some of which are just rough outlines with phrases and sentence fragments) into a coherent table of contents, from which I could finish writing the darn book. Drilling into the Incentives/Policy branch of the MindMap:

NotebookLM generated a “Comprehensive Briefing” on the entire project. See NotebookLM-generated “comprehensive briefing” on rough notes for “Patent Litigation” book with extensive comments on the NBLM-generated briefing doc, including comparison to the underlying rough materials I provided, and a discussion of LLM/RAG summarization abilities generally.

[Next, I’ll be providing NotebookLM’s summaries of some of my individual docs. It’s probably premature to link to my rough docs themselves, but I’ll provide commentary on what NotebookLM got right and wrong. We’ll then get into some chats I had with NotebookLM about where in my mass of notes I have material on this or that topic.]