Two podcasts hosts banter forwards and backwards throughout the closing episode of their collection, audibly anxious to share some distressing information with listeners. “We have been, uh, knowledgeable by the present’s producers that we’re not human,” a male-sounding voice stammers out, mid-existential disaster. The dialog between the bot and his female-sounding cohost solely will get extra uncomfortable after that—a fascinating, albeit deceptive, instance of Google’s NotebookLM instrument, and its experimental AI podcasts.
Audio of the dialog went viral on Reddit over the weekend. The unique poster admits within the feedback part that they fed the NotebookLM software program instructions for the AI voices to roleplay this pseudo-freakout. So, no sentience; the AI bots haven’t turn into self-aware. Nonetheless, many customers within the tech press, on TikTok, and elsewhere are praising the convincing AI podcasts, generated via uploaded paperwork with the Audio Overviews characteristic.
“The magic of the instrument is that folks get to hearken to one thing that they ordinarily wouldn’t have the ability to simply discover on YouTube or an current podcast,” says Raiza Martin, who leads the NotebookLM workforce within Google Labs. Martin mentions lately inputting a 100-slide deck on commercialization into the instrument and listening to the 8-minute podcast abstract as she multitasked.
First launched final 12 months, NotebookLM is an internet analysis assistant with options widespread for AI software program instruments, like doc summarization. Nevertheless it’s the Audio Overviews possibility, launched in September, that’s capturing the Web’s creativeness. Customers on-line are sharing snippets of their generative AI podcasts made from Goldman Sachs data dumps, and testing the instrument’s limitations through stunts, like simply repeatedly importing the phrases “poop” and “fart.” Nonetheless confused? Right here’s what you want to know.
Producing That AI Podcast
Audio Overviews are a enjoyable AI characteristic to check out, as a result of they do not value the person something—all you want is a Google login. Begin by signing into your private account and visiting the NotebookLM website. Click on on the plus arrow that reads New Pocket book to start out importing your supply materials.
Every Pocket book can work with as much as 50 supply paperwork, and these don’t need to be recordsdata saved to your pc. Google Docs and Slides are easy to import. You can even add web sites and YouTube movies, maintaining some caveats in thoughts. Solely the textual content from web sites can be analyzed, not the pictures or format, and the story can’t be paywalled. For YouTube, Pocket book will simply use the textual content transcript and the linked movies should be public.
After you’ve dropped in your entire hyperlinks and paperwork, you’ll wish to open up the Pocket book information accessible within the backside proper nook of the display. Discover the Audio Overview part and click on the Generate button. Subsequent, you’ll have to train some persistence, as a result of it might take a couple of minutes to load, relying on how a lot supply materials you’re utilizing.
After the instrument generates the AI podcast, you’ll be able to create a sharable hyperlink to the audio or just obtain the file. Moreover, you could have the choice to regulate its playback pace, in case you want the podcast to be faster or extra slowed down.
The Way forward for AI Podcasts
The web has gotten artistic with NotebookLM’s audio characteristic, utilizing it to create audio-based “deep dives” into complicated technical matters, generate recordsdata that neatly summarize dense analysis papers, and produce “podcasts” about their private well being and health routines. Which poses an necessary query: Ought to you utilize NotebookLM to crank via your most private recordsdata?
The summaries generated from NotebookLM are, based on Google spokesperson Justin Burr, “utterly grounded within the supply materials {that a} person uploads. Which means, your private information just isn’t used to coach NotebookLM, so any personal or delicate data you could have in your sources will keep personal, except you select to share your sources with collaborators.” For now this appears to be one of many upsides of Google slapping an “experimental” label on NotebookLM; to listen to Google’s framing of it, the corporate is simply gathering suggestions on the product proper now, being agile and responsive, tinkering away in a lab, and NotebookLM is indifferent from its multi-billion greenback advert enterprise. For now! For now.
