Opinion: Nobody needs AI-generated podcasts

In mid-September, Google released Audio Overview, a new feature for the NotebookLM note-taking tool that allows you to create a podcast from two AI hosts for an uploaded file or URL. In the best babble podcast style, two text-to-speech systems then talk to each other about the content of the information provided using a podcast script created with Gemini 1.5. So far, you can't customize the dialog and the system doesn't seem to want to disclose that it's an AI-generated rehash.

An opinion by Philipp Steevens

Philipp Steevens has been with iX since 2022. He is primarily responsible for articles in the fields of data science and AI and is in charge of the magazine's LinkedIn presence.

As I rarely listen to podcasts, the whole thing initially left me cold, although the feature has made a few waves across the internet. But after listening to it for the first time, I can say that I have rarely seen such a useless processing of information – or, in this case, heard it. So far, the whole thing has only worked as an experimental feature in English, but the success, at least in terms of humor, suggests that it will spread to other languages.

Basically, I don't think the idea behind NotebookLM is wrong. You upload documents or enter a URL, a language model from Google summarizes the content, highlights key topics and offers questions that you could ask the source. This can be helpful for a quick overview of longer content or perhaps as a first approach to a complex topic. In principle, the whole thing is not very different from RAG systems that you set up with your own documents. Whether you are in favor of using an AI assistant to further develop the topics is certainly a matter of taste, but it is also similar to established coding with an AI programming assistant.

A gimmick, nothing more

What really bothers me about Audio Overview is the idea that the podcast could somehow help you acquire knowledge from your own documents or external sources. A real podcast is interesting to me for various reasons. On the one hand, there is the classic podcast with well-known people speaking freely or in a structured way on certain topics, sharing their opinions and perhaps even categorizing the topic. You get the impression of a more personal insight into the thoughts and opinions of these people, which makes the podcast feel intimate. On the other hand, there are podcasts on specific topics or news podcasts. Here I appreciate a clear central theme and prepared information that the speakers convey clearly and appropriately. For me, the audio overviews from NotebookLM are a bastard of both categories, which above all reinforces the disadvantages of the podcast format.

Audio Overview's podcast AI stumbles over its own sentences, pauses to find words, and really only makes the very worst puns. If you want a scientific paper or technical documentation summarized, bad style would be a compliment at this point. The style pisses me off because I know that it's not a real human being who has to put the sentence together to express something in the best possible way. The thing is a language computer, it doesn't have to copy the peculiarities and mistakes of human speech one-to-one. I don't expect a knowledge aggregation tool to produce sentences that hint at an opinion or feigned empathy that the system behind it can't feel anyway, let alone understand. The back and forth between the pretend moderators only delays the absorption of knowledge for me. Instead of getting condensed knowledge, I have to listen to what feels like 30 seconds of babble for every 15 seconds of information snippets.

Google itself still describes the Audio Overview function as an experiment. For me personally, I would declare it a failure. I would listen to an audio overview feature where you can turn off the unnecessary banter between the simulated hosts again for a new opinion. Until then, I'd rather keep working with text.

(pst)