I had been hearing about this NotebookLM thing for a few days. Someone at the Open Ed conference said it was impressive. Then CogDog gave us a challenge to listen to a nearly two hour recording. While I only made it through a few seconds of that, I did decide to try it out myself. Would that be an example of the backfire effect?
Since I was just in Providence, I thought it would be appropriate to feed the Deep Dive hosts some Lovecraft poetry. I picked Nemesis because I liked the title. I only gave it the text, not the title or author, and it recognized it as a poem by an unknown author. In a follow-up attempt I stripped out all the line breaks to see if it would identify it as poetry, which it did, and it also described it as “lovecraftian.”
I can see how people are impressed at first, because it sounds like a reasonable facsimile of a podcast. But it also reminds me of how people were impressed with ChatGPT at first. Sure it’s coherent English, but give it a second look and you’ll notice it uses a lot of words to say very little. The banter between the speakers, who I want to call Robota and AutomaTom, may be impressive to people who aren’t ordinarily repulsed by such stuff, but really it ads only fluff.
Their conversation does reference and offer some interpretation of specific parts of the text. It also brings up the topic of eternal recurrence, which is not explicitly mentioned, yet present throughout the text. That sounds a bit like an insight. That’s not something I would expect from an AI. I suspect that people have done literary analyses of the poem before, so the “eternal recurrence” phrase is statistically correlated with word patterns from the poem. But still, it’s a term I didn’t know before, so I learned something.
I had the thought that I might mash up the deep dive with a reading of the poem by interweaving the two audio files. I may still do that, but in the process of looking for a recitation, I came across this eldritch nightmare:
Now I can’t read the poem without that rhythm and melody in my head. HP would be proud.
Brilliant toss there, Paul! I was pondering what to toss in the bin that would be ridiculous to hear them exhort about. I still have to admit the conversation pattern is amazing- the “you know”the insertion of “like” and some turns of phrases “like a gothic playlist cover come to life”.
It’s notable how uncritical Tom and Robert’s are, it’s always got a sycophantic “this is so call” attitude for anything.
Well played, sir!
One of my coworkers fed it her report on how ExLibris isn’t working properly with Banner, and the results were impressive. I suggested she submit the podcast to our boss.
While not a “deep fake” as such, one could use the podcast as evidence in a Trumpian like over-generalized “everybody’s sayin’,… everybody’s talkin’,” and dumped whole as an argument for or against any topic. Like a debate club, Pro/Con on any topic just give it the essay to NotebookLLM-ify.