NotebookLM is scary good if limited for now

You should check out Google’s NotebookLM, in particular its podcast creation feature. I gave it two assignments, the first a general discussion of Schopenhauer’s philosophy, and the second, my wildly popular and earth-shattering, and in all fairness mildly interesting article The Web of Law, to summarize and discuss in a podcast format. The language model chews over the documents you provide (including any podcasts, videos, etc. as well as docs and pdfs you have uploaded) and then spits out an 8 minute or so discussion by two articulate, artificial Gen X or Z types of the topics. For Schopenhauer I gave it his Wikipedia page and two long internet encyclopedia articles, and for my article I gave it my article.

The results were remarkable though not terribly thorough analyses of both. The psychological impact of hearing your thoughts discussed by two smart sounding individuals, even though they don’t really exist, was striking to me. It was a new way to get a dopamine hit. On reflection, however, I thought its discussion of my article missed some of its profound insights (or if you insist, its fair points). Still, their discussion was brief, kinda funny, and made at least some of the points I made in the article. There was a fair amount of verbal padding to keep the discussion moving, but it did actually, you know, keep the discussion moving.

The discussion of Schopenhauer focused on the key point of the Will in S’s philosophy, which is fair enough. It began with a discussion of how S was a big supporter of animal rights and loved his poodles, as well as neglecting some notorious examples of S’s misanthropy, as if in an attempt to convince a young, skeptical audience that it was ok to learn something of this philosopher. There was little about S’s relationship to Kant or to later thinkers or artists. A lot of emphasis on S’s love of music and the legitimacy of aesthetic experience, which is fair. All in all, a decent 8 minute introduction for non-major undergraduates, I’d say.

But remember, this is just a machine doing this! And it’s new. If students did these podcasts as an assignment, I’d give them a B, which is not bad. These applications will get better over coming years. Heck, over coming months, the way things are going. Presumably, NotebookLM could produce an equally informed discussion about whatever horribly complicated topic in fluid mechanics or plant genetics you are qualified to judge and not just the soft social sciencey topics I came up with. If we don’t blow ourselves up or go full socialist, this sort of model or whatever you want to call it, will likely become a powerful tool and imaginably even intelligent companions in years to come.