JamesCaruso
Staff member
I started playing with Google Notebook LM for the first time last weekend. For those unfamiliar with it, you create a “notebook,” upload sources (I think up to 50) that can be web links, PDFs or other file types, copy/paste text, or even YouTube videos. You can then interrogate the information via chat, or get one click output in the form of outlined notes, an “FAQ” document, or a “study guide” for studying via active recall. But by far the most incredible feature - the type of amazing “wow factor” level-jump for consumer AI that ChatGPT was - is the ability to synthesize all of the sources into an audio file that takes the form of a podcast, with two virtual “hosts” discussing the topic. The tone is conversational, and the voices are human-like enough to be pleasant, with the inclusion of fillers like “um,” “like,” “you know,” and even a simulated chuckle here and there. A newer feature even allows you to stop the conversation, ask a question, and get an answer from the hosts.
So how is this different than ChatGPT, Claude, etc.? Well the podcast generation is certainly one big one. Others: Notebook LM chat responds only relative to the sources that have been provided. The user interface and visual metaphors are a bit different - there are three main windows - sources, chat, and the deliverables (audio file, FAQ, study guide, etc.) The landing screen shows each of your “notebooks,” so you can organize your projects and/or learning topics. It feels like a better way to learn and manage information; it’s more “organized” and structured than ChatGPT. There is also an ability to add your own notes into the mix, and I think you can then have them treated as a source or not.
I tried a couple of weather related things and generated podcasts to experiment. First, I gave it a link to my local NWS forecast office AFD. It was interesting to hear the podcast “hosts” do an intro kind of joking about how the AFD is such an in-depth forecast. Next, I tried to give it a link to the SPC convective outlook for El Reno day in 2013. It was unable to use the link; perhaps SPC has some security that blocks applications like this from accessing its content? So I copy/pasted the outlook text in. I expected the podcast to just be an objective verbatim discussion of the content. Instead, it recognized it as a past event, and talked about it in that context. Although supposedly Notebook LM only looks at the source material, the podcast explained certain concepts and terminology used within the outlook, so I’m surprised by that. Not in a particularly comprehensive way, but still pretty cool. It also added a conclusion to the effect of “At the risk of being a spoiler, did this forecast turn out to be accurate that day?” The other host responded that it did, although there were no specifics about the tornado or the TWISTEX tragedy.
I think this is a fanstastic tool for learning (i.e., upload a bunch of research papers on a topic; copy/paste chase reports to synthesize a past event; give it info on a chase day - convective outlooks, area forecast discussions, “event” posts from ST - and let it synthesize the information, or listen to the “podcast” in the car as you drive toward the target. I think it will be a great way for my son to learn when he chases with me, if the podcast includes explanations about certain technical terms or parameters.
If anyone has experimented with this, I’d be curious to hear about your experiences and other use cases.
Here are a couple links to get familiar with Notebook LM if you are new to it:
blog.google
blog.google
EDIT: In the section above about differences vs ChatGPT, I added a point about note-taking.
So how is this different than ChatGPT, Claude, etc.? Well the podcast generation is certainly one big one. Others: Notebook LM chat responds only relative to the sources that have been provided. The user interface and visual metaphors are a bit different - there are three main windows - sources, chat, and the deliverables (audio file, FAQ, study guide, etc.) The landing screen shows each of your “notebooks,” so you can organize your projects and/or learning topics. It feels like a better way to learn and manage information; it’s more “organized” and structured than ChatGPT. There is also an ability to add your own notes into the mix, and I think you can then have them treated as a source or not.
I tried a couple of weather related things and generated podcasts to experiment. First, I gave it a link to my local NWS forecast office AFD. It was interesting to hear the podcast “hosts” do an intro kind of joking about how the AFD is such an in-depth forecast. Next, I tried to give it a link to the SPC convective outlook for El Reno day in 2013. It was unable to use the link; perhaps SPC has some security that blocks applications like this from accessing its content? So I copy/pasted the outlook text in. I expected the podcast to just be an objective verbatim discussion of the content. Instead, it recognized it as a past event, and talked about it in that context. Although supposedly Notebook LM only looks at the source material, the podcast explained certain concepts and terminology used within the outlook, so I’m surprised by that. Not in a particularly comprehensive way, but still pretty cool. It also added a conclusion to the effect of “At the risk of being a spoiler, did this forecast turn out to be accurate that day?” The other host responded that it did, although there were no specifics about the tornado or the TWISTEX tragedy.
I think this is a fanstastic tool for learning (i.e., upload a bunch of research papers on a topic; copy/paste chase reports to synthesize a past event; give it info on a chase day - convective outlooks, area forecast discussions, “event” posts from ST - and let it synthesize the information, or listen to the “podcast” in the car as you drive toward the target. I think it will be a great way for my son to learn when he chases with me, if the podcast includes explanations about certain technical terms or parameters.
If anyone has experimented with this, I’d be curious to hear about your experiences and other use cases.
Here are a couple links to get familiar with Notebook LM if you are new to it:
8 expert tips for getting started with NotebookLM
NotebookLM is an AI-powered tool for more deeply understanding something — we asked a Google expert to give us a few tips on how to use it.
NotebookLM gets a new look, audio interactivity and a premium version
NotebookLM is introducing new features, and a premium version called NotebookLM Plus.
EDIT: In the section above about differences vs ChatGPT, I added a point about note-taking.
Last edited:
