Focus of the letter 34

In the readings of letter 27, I quoted an article from Libération on an AI podcast that produced 11,000 episodes a day by sucking up articles from the local press, a few minutes after their publication with the same facts, the same structure and the same turns of phrase. The story of Bill Adair, told by Libération, the journalism professor who stumbled upon a podcast episode dedicated to one of his students, generated from an article published a few minutes earlier, sums up well what this case of AI content reuse now has that is both banal and disturbing. This example is not just an anecdote: the article cites as an example that in May 2025, the Paris court ordered the blocking of a site that produced 6,000 articles per day by looting about forty French newspapers.

This week, while testing NBot, Dailyscope and Podcasty, I realized that what seems to be an industrial infrastructure is now within anyone’s reach, in a few clicks and without technical skills. Dailyscope aggregates and compares international press sources every day. NBot builds a watch on a given subject and can generate a daily podcast. Podcasty transforms any article into an audio episode in 57 languages. What Daily News Now! has industrialized, these three tools make it accessible to everyone.

In letter 29, I shared Peter Gostev’s Bullshit Benchmark which shows something simple and a little dizzying: ask the models questions with correct language but basically invented, and most answer with confidence and detail. Verisimilitude takes precedence over meaning, what looks like a good answer passes for a good answer. Daily News Now! works on the same principle but on the scale of a complete editorial chain: what looks like local news, with a city name, the name of a student, a duration, a title, passes for local information.

The audio format adds something special to this equation, and it’s not just a question of listening comfort. Playback allows you to go back, annotate, pause reflexively: you reread, you stop, you check. Listening follows a flow that you don’t control in the same way, and the podcast format adds something to it: the regularity of the episodes, the subscription relationship, the continuity. You don’t consult a podcast like you check a piece of information, you follow it without often going back to a passage of the episode you have listened to or to a previous episode.

Bill Adair says it clearly in the article in Libération: what has changed is not the principle of looting, which is not new, it is its scale. A human was limited by his ability to process, this limit has disappeared.

These three applications show us a new swipe, the same one I used to describe my Echo game . We set up a watch because it’s convenient. We listen to the podcast because it’s there. We don’t check because the format doesn’t lend itself to it and because it looks like what we expected.

The question is not so much “how to detect what is generated” as “do we still maintain the habit of asking who chose, why, and what is not in the flow?”

This is a question that can be asked in front of any information, generated or not. But with these tools, the ease of forgetting this question is made a little easier than before…