A look at Google NotebookLM's Audio Overview feature, which was built by Google Labs and has gone viral for its AI-generated “deep dive” podcasts with two hosts
Thoughts on a new feature at NotebookLM that turns your sources … Roger Dooley / Forbes : How To Create An AI Podcast About Anything In Seconds With NotebookLM Scott Rosenberg / Axios : Google's AI podcast hosts draw crowds Ian Carlos Campbell / Pocket-lint : Google NotebookLM made an entire podcast from my notes in just a few clicks Hard Fork on YouTube : NotebookLM Blew Our Mind | Interview Melissa Heikkilä / MIT Technology Review : People are using Google study software to make AI podcasts—and they're weird and amazing Carl Franzen / VentureBeat : Google NotebookLM leader says more controls coming for AI generated podcasts Wired : How to Generate an AI Podcast Using Google's NotebookLM X: Ethan Mollick / @emollick : Our research a year ago found that people stopped fact checking the AI when it got good enough, and juat took what it said as right (even if it wasn't). I think that line has firmly & permanently been crossed for many text summarization applications. Ethan Mollick / @emollick : NotebookLM's podcasts seem to hallucinate at a similar level to other Gemini 1.5 text summaries (so greatly reduced compared to previous LLM models, but they happen) I suspect that we see less complaints, however, as the audio format lulls us & also make noticing errors harder Andrej Karpathy / @karpathy : Over the last ~2 hours I curated a new Podcast of 10 episodes called “Histories of Mysteries”. Find it up on Spotify here: https://open.spotify.com/... 10 episodes of this season are: Ep 1: The Lost City of Atlantis Ep 2: Baghdad battery Ep 3: The Roanoke Colony Ep 4: The Antikythera [image] Andrej Karpathy / @karpathy : Deep Dive is now my favorite podcast. The more I listen the more I feel like I'm becoming friends with the hosts and I think this is the first time I've actually viscerally liked an AI. Two AIs! They are fun, engaging, thoughtful, open-minded, curious. ok i'll stop now. Amber Mac / @ambermac : These are not real people doing a podcast chat about my newsletter intro. Instead, I dropped my 200-word essay text into Google's NotebookLM (a virtual research assistant) and chose the little button at the top that says - Deep dive conversation: Two hosts (English only). Now [video] Dave Troy / @davetroy : Google's NotebookLM is by far the most actually useful AI product of the current generation, for people that actually need to augment their own capacity. Being able to process multiple large and complex sources and pull out key details (names, organizations, key themes, etc) is Steven Johnson / @stevenbjohnson : Love that you're doing this, @omarsar0! Sounds like a great intro (and beyond) to the product. Elon Musk / @elonmusk : @karpathy Interesting Eric Jorgenson / @ericjorgenson : @karpathy How do you fact-check generations like this? @cnbctechcheck : We tested out $GOOGL's viral NotebookLM tool using CNBC and Wall Street's favorite study tool: an analyst note. Powered by Google's most advanced AI model, Gemini 1.5 Pro, could this be the next killer app for generative AI? @dee_bosa reports [video] Andrej Karpathy / @karpathy : @EricJorgenson Good q worth noting that the Wikipedia page is in the context window when the pod is generated, this increases the accuracy a lot provided original material is ok. It's when you try to “zero shot” prompt straight up you'd expect a hazy recollection, quite possibly hallucinations. @omartosca : Many people have praised NotebookLM for their approach to genAI podcasting, even the UI for researching, but they all forget to praise the model behind. Gemini is more powerful and impressive than people think, they just need to get out of the gpt and Claude bubbles. Elvis / @omarsar0 : 🎓Introduction to NotebookLM Very excited to announce @dair_ai's first FREE course, Introduction to NotebookLM! As an AI educator, researcher, and engineer, NotebookLM unlocks a wide range of use cases. It's not just the podcast generation feature (which is amazing), it's [image] Steven Johnson / @stevenbjohnson : Wow. Such a great case study in why we designed NotebookLM from the beginning around curated sources It's basically the foundation of the product: curate the information you need the AI to have expertise in—and then transform it into whatever form you want. Tae Kim / @firstadopter : Congrats to Google Labs and the NotebookLM team for creating one of the most compelling AI products yet. Great stuff. It offers a glimpse of AI's future with Gemini's longer context window capabilities and unlocking the siloed knowledge inside corporate internal databases [image] See also Mediagazer
Context & Ripple Effects
NotebookLM first added Audio Overview as a way to turn supplied documents into a two-host discussion; early coverage emphasized how the feature could summarize and explain source material in audio form. The subsequent viral response suggests the product’s appeal lies not only in summarization, but in making research outputs feel immediately consumable.
The feature is powered by Gemini 1.5 Pro and was developed in Google Labs, while NotebookLM leadership has indicated that more controls are coming. That combination makes it a notable test of whether polished, source-grounded generative formats can pull AI from a text prompt interface into a repeatable research workflow.
First-order effects
- NotebookLM users and creators can turn their own notes and source collections into conversational audio, lowering the effort required to make an audio-oriented explainer or study aid.
- The viral “Deep Dive” format gives Google a tangible consumer-facing use case for Gemini, while putting pressure on NotebookLM to add the controls its leadership has signaled as usage broadens.
Second-order effects
- AI research-product rivals must compete on output format and usability, not just model quality: the earlier document-to-audio launch established a workflow that this broader attention now validates.
- Creators may test audio versions of existing research and notes before investing in conventional production, increasing the supply of synthetic explainers while making provenance and user control more important.
Third-order effects
- If source-grounded audio becomes a standard research output, AI tools will increasingly compete as interfaces that transform a user’s material into multiple media forms rather than as standalone chatbots.
- The format could make synthetic audio commonplace in knowledge work, but its durability will depend on whether providers can offer enough editing, attribution, and source transparency to distinguish useful synthesis from disposable output.
The trend: Generative AI products are shifting from answering prompts to repackaging user-provided material into workflow-native, multimodal outputs.