NotebookLM is an Amuse-Bouche
That's just me being all fancy-pants, calling NotebookLM an amuse-bouche. An amuse-bouche, Google tells us, is "a small savoury item of food served as an appetizer before a meal".
And NotebookLM, of course, is the cool new technology that is able to generate podcasts based on anything that you provide as an input.
And when I say anything, I am not exaggerating:
https://twitter.com/deliprao/status/1840959756661383653
(H/T: TheZvi's newsletter on AI, which is what this post is based on)
TheZvi tried the podcast feature out, and came away somewhat impressed:
"In some ways I was impressed. The host voices and cadences are great, there were no mistakes, absurdities or factual errors, everything was smooth"
But the key word is "somewhat":
"In terms of being an actual substitute? Yeah, no. It did give me a good idea of which ideas are coming across ‘too well’ and taking up too much mindspace, especially things like ‘sci-fi.’ I did like that it led with OpenAI issues, and it did a halfway decent job with the parts it did discuss. But this was not information dense at all, and no way to get informed."
And I agree, with one minor quibble. For all the inputs that I have given to NotebookLM, it has been able to generate fun, diverting podcasts that sound peppy, seemingly informative, and definitely end up giving a flavor of what is contained within the input. But by no stretch of the imagination is it a substitute for the original thing. Not even close.
But I'd argue that you shouldn't be optimizing for a substitute. Is it a useful complement to your input? And here, your mileage may vary, sure, but I'd think a lot more people would agree that it is a useful complement.
Say, for example, you are reading about The Problem of Social Cost for the first time. That's a long review article to read, at least by first year undergraduate standards. Not one that needs you to know much (any) mathematics, but not a light read, not by any stretch of the imagination.
As with everything else in life, you may want to begin by approaching the problem in lots of different ways. Maybe you want to watch a YouTube video about the topic. Maybe you want to read the Wikipedia article. Maybe you want to read the original paper. And hey, maybe you want to listen to two people talking about some or all of these things.
None of these things are a substitute for sitting with the original article and spending time with it. Perhaps they'll never be an acceptable substitute to reading the whole thing. But all of them are useful complements to the act of learning and internalizing the paper.
And at the margin, having an additional arrow in your quiver is a wonderful thing.
The mistake would be to say that you only need listen to the podcast, and choose to ignore the original material. That's selling yourself short.
But for at least some people, it would also be a mistake to say don't bother listening to the AI-generated podcast. If it raises interest in reading the original, if it increases your ability to better engage with the original, and if it helps you think about the original in better ways, than it is, on balance, a Very Good Thing.
Depends, as always, on how you use the technology. A starter can never (well, ok, should never) replace the main course. But that doesn't mean the starter is useless. Some of us may choose to skip it, sure, but for a lot of us, there's a lot of joy and help to be found in the complement.
So, as I was saying:
