Recently NotebookLM started to pop up at work as part of our Google Enterprise subscription.
We have a SaaS committee that reviews tools, tries to consolidate licensing, etc – and so this was brought up more as a curiosity I guess. I saw a demo, didn’t really think much of it – maybe the demo was too short, not sure.
However, then a test project was suggested and I thought OK, time to see what this is about.
To start I found a Udemy course by Scott Duffy (link at the bottom) – a quick one hour overview. To be honest if we didn’t have a Udemy business account, not sure I would have bought this class – but if you are on a personal or team plan, it might be worth an hour of your time to watch if you have no exposure to this.
Of interest to me was one of his examples where NotebookLM was used to look at a github repo and then make an Audio Overview – essentially it generates an audio file with two AI voices interacting conversationally about the sources you’ve added. Basically a podcast.
A quick step back – unlike chatGPT, Gemini, etc – NotebookLM allows you to add “sources” to a notebook and it uses that as the basis for its LLM in that scenario. So in my case, I added my repo and then asked it some questions and had it generate the podcast.
The first pass was OK. It looked at my repo, the comments and correctly deduced that this was a fun sample project that I did with my son 4 or 5 years ago to learn Python and get progressively more complex. It’s a text dungeon game that evolved a bit, and was never meant to be anything that a way to work on a fun little project together.
Once you create an Audio Overview, you can delete it – and then a Customize button appears where you can guide the conversation of the podcast. I wanted it to be a little more detailed about the actual code so I added links to a few specific files in the repo and asked it to focus on those. The resulting audio file was pretty good – a decent conversation between the AI’s about the game and how it worked. It had some weird pronunciations, but overall it was pretty cool.
From there you can do all sorts of things like generate a briefing doc, study guides if you want to make a quiz plan for it, FAQs, and a timeline of the project.
NotebookLM also has a chat interface like chatGPT and Gemini, but it can only really answer questions about the sources you’ve added to a given notebook. It will let you know if it can’t do something, and as long as you have space (I think you can add up to 50 sources as of now) you can ask again, and keep iterating.
It can do a lot of other things as well – but for me it was a first quick afternoon experiment to see what it is.
What I’d love to see next is how to interact with it programmatically. Some automation around its sources, can you pre-define some questions or reports on the sources, etc. More on that as I do a bit more with it.
You can learn more about NotebookLM here: https://notebooklm.google.com/
The Udemy Class I mentioned is here: https://www.udemy.com/course/notebook-lm