^ this is important.
Otherwise you may very well be missing anything really surprising or novel.
See for example https://www.programmablemutter.com/p/after-software-eats-the... , an experience report of NotebookLM where
> It was remarkable to see how many errors could be stuffed into 5 minutes of vacuous conversation. What was even more striking was that the errors systematically pointed in a particular direction. In every instance, the model took an argument that was at least notionally surprising, and yanked it hard in the direction of banality.
On the other, Google might not have done much to upgrade the podcast feature since them.
Sometimes I'll take deep research output and listen to it too that way.
This somewhat makes whole NotrbookLM less useful, but still.
Having said that I absolutely hate the audio format, I only used it when I had to drive or when I swam lanes. But these days I do neither.
For example, I can give it 8 papers on best practices in online marketing, it will turn it into a 20 minute podcast.
There are errors, but also with real podcasters.
Or before! Either is mandatory to actually learn the content.