Oct 8, 2024
In mid-2023, Google introduced “Project Tailwind,” an AI-powered research platform. Over the past year, the product received numerous updates and was rebranded as NotebookLM. The latest update allows you to upload files and link to online resources, have it summarize the source material, create learning resources, and produce a podcast with two hosts discussing the source material. The latter feature attracted significant media attention—and the results are truly wild. From study guides, summarizing complex technical documents and research white papers, to book summaries and much more.
Which brings me to today’s newsletter—I had NotebookLM generate a podcast of my book, Disrupt Disruption, and the results are fascinating (in many ways). Listen here:
There is a lot to unpack here. First, it’s truly surreal and impressive hearing two AIs discuss my book (in a rather engaging way) within five minutes of me uploading a PDF of my book to NotebookLM. On the other hand, something feels off. Jeffrey summed it up perfectly in a message he sent when the team first listened to the podcast:
It’s certainly impressive—both the summarization capability and the naturalistic speech (the filler words, talking over each other a bit, and each speaker having some specific vocal tics). And it’s impressive to think about how far that’s all come in the last year or so.
There are still some weird inconsistencies and some things that are too consistent and repetitive, but I get that all of this will also likely be improved soon.
But I can’t help finding the whole thing weirdly depressing—the idea that we’re listening to conversations that didn’t happen between people who don’t exist and don’t have any real POV on the book that comes from lived experience or personal enthusiasm, etc.
Like, the blandest possible content about a really interesting book by a really interesting human with a very distinctive voice and POV as a writer! 🤯🤯🤯”
This perfectly captures my own ambivalence: As impressive as this tech is, it comes at the cost of flooding the world with AI slob—quite possibly to an extent where we devalue content so much (as there is so much of it) that we will just stop caring.
Finally, two more takes on NotebookLM: one from someone who hacked NotebookLM’s AI podcast hosts to realize they aren’t real, and the other a profound conversation that ensued when NotebookLM was fed a long document entirely made up of the two words “poop” and “fart.”
P.S. I’d love to hear your thoughts on the podcast. 🤗
@Pascal
