Google has just enhanced their Notebooklm product, a personalized AI research assistant, with a truly magical feature: the ability to generate a ~10-minute ‘podcast’ based on any text you provide, using their most advanced LLM model, Gemini Pro 1.5.
I just tried it with the article I wrote about the LLM-based scheduling assistant I’m working on. The result is pretty fascinating and certainly entertaining! Give it a listen and judge for yourself.
The generation process is very straightforward.
After creating an account at Notebooklm, create a new Notebook:
Then, select the sources to be used to generate your podcast (you can add up to 50 per Notebook):
I selected Website and entered the URL of my article:
Immediately, a ‘Notebook guide’ is generated, complete with a summary of the article, suggested questions for further exploration, and buttons to generate an FAQ, Table of Contents, Study Guide, and… an Audio Overview:
Now, simply click on Generate,
and in just a few minutes you will get a ~10 minute “Deep dive conversation” between two English speaking hosts!
It’s really that simple.
The product is still fairly basic though: you can’t change the voices of the hosts, adjust the tone of the conversation, set its length, etc. It would also be great if you could prompt the system to generate a more personalized podcast; for example in my case, it would be interesting to listen to a host-guest format, where the host interviews the author of the article, with their own voice, as a guest.
Nonetheless, the result is impressive for a first iteration, offering an engaging and convenient way to digest complex information. It’s also a good example of how LLMs combined with speech generation can be productized to deliver useful and practical solutions.
Share this post