I want to know why, not just get an answer
Nutrition was the first example I picked for this NotebookLM test. As an athlete, I want to understand why I'm doing something. I don't have a big staff around me researching everything on my behalf.
I tried three subjects in the video: nutrition, AI agents and game theory. Then I used a different output for each: an audio conversation, a video overview and a slide deck.
The useful part for me was being able to collect material, ask questions about it and try different ways of going through the same subject. That isn't the same as proving every answer correct.
Nutrition research I could ask questions about
For the first notebook, I collected sources about nutrition and asked a question using my own height and weight. The answer was generated from the material in the notebook.
I also opened one of the source links in the recording. That's a step I'd keep doing. Having a source attached makes an answer easier to investigate; it doesn't mean I should accept it without reading.
The video title talks about replacing a nutrition coach. The demonstration doesn't establish that. This was a way for me to explore information, not a replacement for qualified advice or an individualized nutrition plan.
The nutrition example begins around 1:30.
Asking a question inside the audio conversation
Have you ever listened to a podcast and wanted to ask the hosts something?
That's the part I was excited to try. While the generated audio discussion was playing, I joined in and asked a question about the nutrition material. The AI hosts responded to me.
You can see that exchange from 8:30. It's an interaction with generated hosts, not a consultation with a person.
AI agents as a video, game theory as slides
For the second notebook, I asked for material explaining AI agents. I then generated a video overview and played part of it.
For the third, I asked to learn the core principles of game theory in simple language. I generated a slide deck and looked through the result.
I liked being able to change the format depending on what I wanted to do with the information. Listening, watching and revising slides are different experiences. The recording shows those outputs, not a measured improvement in how much I remembered afterward.
How I'd try it on another subject
I'd start with one question rather than an entire field. Add sources I can identify, ask the question, and open the references behind anything important. Then choose the output I'd actually use.
One correction to something I said in the recording: I called Reddit owned by Google. The announcement was a partnership giving Google access to Reddit content, not an acquisition. That aside wasn't a reason to trust a source.
Watch the three demonstrations to see the questions, generation steps and outputs.
Resources
- NotebookLM: the product shown in this recording.
- Google's Audio Overview documentation: current guidance. Names, availability and controls can change after a recording.