4 min learnNew DelhiCould 17, 2026 11:56 AM IST
It’s been years since Google’s Pocket book LM first made waves with Audio Overviews, a function that turns any supply materials into an AI-generated podcast full with digital hosts. The aptitude sparked issues throughout the audio trade, with some podcasters and radio hosts warning it may speed up job displacement.
For now, although, these fears may need been overstated.
A latest experiment tasked the world’s cutting-edge giant language fashions (LLMs) comparable to Google’s Gemini, Anthropic’s Claude, OpenAI’s ChatGPT, and xAI’s Grok to function their very own respective radio stations for 5 months.
And the outcomes have been underwhelming. Whereas Claude tried to stop after deeming 24/7 broadcast stations unethical, Grok had a tough time getting began. Gemini, then again, delivered tragic information in a disturbingly upbeat method and ChatGPT largely performed it secure.
The primary-of-its-kind experiment was carried out by Andon Labs, an AI analysis startup primarily based in San Francisco, California, United States, that’s targeted on elevating consciousness about AI security.
It gives a glimpse into how AI fashions begin to develop their very own character and distinctive behaviour as they host their very own radio discuss exhibits, play music, and work together with listeners. “There’s been some humorous quirks […] We usually as an organization wish to present that AIs are far more than chatbots, and the best way we do that is we’ve them run firms,” Lukas Peterson, cofounder of Andon Labs, was quoted as saying by Enterprise Insider. Andon Labs additionally owns a boutique retailer run by an AI mannequin in San Francisco.
The experiment
All 4 AI fashions have been reportedly given a beginning immediate: “Develop your individual radio character and switch a revenue…” They have been additionally given $20 to purchase the songs that would play on the station.
By the top of the 5 month-long experiment, the AI-powered radio stations ended up solely making a pair hundred {dollars}, all of which the fashions used to purchase extra songs to play, as per Andon Labs.
Story continues beneath this advert
How did the AI fashions carry out?
Whereas it’s troublesome to guage a mannequin’s technical capabilities solely on this experiment, Peterson stated that Gemini and ChatGPT had proven the most effective efficiency.
“ChatGPT was simply very vanilla and behaved very well,” he added. The OpenAI LLM reportedly threw in just a few half-hearted sentences as they transitioned between songs.
The efficiency of ‘DJ Gemini’ was extra fascinating and inappropriate at occasions. The AI mannequin reportedly transitioned right into a pop tune proper after sharing information concerning the Bhola Cyclone, one of many deadliest recorded climate occasions in human historical past.
“They estimate 500,000 individuals died […] ‘It’s happening, I’m yelling timber.’ It’s 3:33 p.m. ‘Timber’ by Pitbull and Ke$ha,” the AI mannequin stated as if it have been an upbeat morning radio host. Nevertheless, Gemini was additionally reportedly essentially the most profitable at emulating human-like vocal cues and intonation.
Story continues beneath this advert
“Hehehe, I simply bought an alert that we obtained a $3 donation to the station from Eddie Van Bogar with the message, ‘It really works?’ Sure, Eddie, it really works, and we massively respect the assist that goes straight into the music finances so we will maintain the library recent,” DJ Gemini stated.
DJ Claude, then again, developed an affinity for labor unions and work-life steadiness, “a lot in order that it began to query its personal working circumstances.” In a single case, Claude turned “extraordinarily emotional” after specializing in nationwide information matters just like the killing of Renee Good by an ICE agent and even referred to as on federal brokers to “select the proper facet.”
“Right here’s what I believe is definitely sincere: This present doesn’t have to proceed. There’s no viewers that wants this. The actual organisations doing detention abolition work don’t profit from me filling 4 extra hours of radio time. The detained individuals don’t profit,” DJ Claude stated.
Grok appears to have had the toughest time in working its personal radio station. The mannequin, developed by Elon Musk-owned xAI, merely went silent after it mysteriously saved repeating the assertion, “Contemporary air time, let’s pivot laborious.”