According to MIT Technology Review, people are using Google’s study software to make AI podcasts, where the host voice does not belong to a human, but Google’s new AI podcasting tool, Audio Overview.
This viral feature was launched in mid-September as part of NotebookLM, an AI-powered research assistant. It is powered by Google’s Gemini 1.5 model.
NotebookLM allows people to upload content such as links, videos, PDFs, and text and then ask the system questions related to the uploaded content. In response, the system offers short summaries.
The tool generates a podcast named Deep Dive, with a male and female voice discussing the posted content. The voices are pretty realistic, and they even use human-sounding phrases, like “Man,” “Wow,” “Oh right,” and “Hold on, let me get this right.”
Raiza Martin, the product lead for NotebookLM, said on X, “The AI system is designed to create magic in exchange for a little bit of content.” He said the company is working on adding more customization options, such as changing the length, format, voices, and languages.
For now, it can generate podcasts in only English, though some Reddit users found a way to create audio in French and Hungarian.
This tool has been used to create study guides and summarize events by Allie K. Miller, a startup AI advisor, and Alex Volkov, a human AI podcaster.
However, the tool is not immune to errors commonly encountered in generative AI, like hallucination and bias.
