Google's NotebookLM is a Great New GenAI Tool
You can get Gemini to answer questions based on your own notes or books
Google has released NotebookLM a new free tool and it is something you should definitely check out — because it allows you to use all the power of Gemini 1.5 (Google’s ChatGPT equivalent) but with your own collection of documents.
Google describes it as:
NotebookLM is your personalized AI research assistant powered by Google's most capable model, Gemini 1.5 Pro. Collaborate with a virtual research assistant. When you upload the documents that are central to your projects, NotebookLM instantly becomes an expert in the information that matters most to you.
For example, over the years, I have collated over 22MB of notes that I use for my teaching, my two substacks, YouTube channel and more. What I’ve now done is upload all of that into my NotebookLM and now I can talk to this NotebookLM as if it is a research assistant and ask it to find information, generate lists, find supporting arguments, find citations to relevant references—all based on my own notes.
In fact, it is better. Last week, I was working on a script for a video on the topic “Stress and Your Heart” based partially on my notes but also heavily based on the material from the book “Why Zebras Don’t Get Ulcers” by Robert Sapolsky. For this, I just uploaded the PDF of this book1 into my NotebookLM and then started talking to it as if it were an assistant that has read all my notes and the book and was helping me with research for the script.
Here are some examples. In every screenshot, my question is at the top right, and NotebookLM’s answer is below it.
Here, I’m asking for an overview based on my notes (all 22MB of them) and the book (212 pages)! You’ll notice that the text it has produced has references to the original material (the numbers in little grey circles). Clicking on them will take me to the section of my notes or the book that contains more details related to the preceding sentence.
In the next one, I am giving it a statement and asking it whether that statement is supported by evidence anywhere in my notes/book. You can see that it has answered in the affirmative. and given me the specific ways in which the notes/book support my claim.
One thing that you really, really need to drill into your brain is that LLMs are liars. You can’t trust the output of an LLM. Thus, to make sure that my claim in the screenshot above is actually supported by the notes/book, I clicked on the reference (the number I’ve circled in red). That took me to the exact paragraph containing the details I needed, highlighted in purple. I could quickly read that and confirm that this is indeed true and not a hallucination of the LLM.
I spent much more time with NotebookLM while writing the script. I won’t reproduce everything else here, but I’ll make these points:
It significantly sped up the research I did while writing the script. Doing the same thing without NotebookLM would have taken much longer. I would have used keyword searches over my notes and the book and that just doesn’t work as well. Because you will notice, I want to search by fairly complex concepts that are referenced in the material to in many different ways. More importantly, I wanted only those chunks of the material where multiple concepts intersected (“stress” and “heart” or “parasympathetic nervous system” and “shut down”). Look at the purple source section: you’ll notice that it does not have the phrase “parasympathetic nervous system” or the phrase “shut down” anywhere in it!
Technically, I can upload PDFs into ChatGPT, and talk to the PDFs. It is supposed to give similar functionality but it doesn’t work as well. Most importantly, it doesn’t give direct links to the source material from which each sentence/claim has been taken. And it definitely does not make the whole process so smooth.
I expect that I will be using NotebookLM quite a lot from now on for all research-based articles and scripts I write
I think I’ve just scratched the surface of what is possible. I expect to discover more cool ways of using this technology as time goes on.
If you can think of other interesting ways of using NotebookLM, please let me know in the comments
And, as usual, please forward this post to anyone you know who would benefit from this:
I have paid for both, the Kindle version of this book as well as the Audible version. So I did not feel guilty of downloading the PDF of the book from libgen.