It’s been several years since Google’s Notebook LM initially drew attention with Audio Overviews, a tool that converts any reference material into an AI-produced podcast featuring synthetic hosts. The feature triggered anxiety throughout the audio sector, with certain podcasters and radio presenters cautioning it might speed up workforce displacement.
At least for the moment, however, those concerns may have been exaggerated.
A new trial assigned the world’s most advanced large language models (LLMs) including Google’s Gemini, Anthropic’s Claude, OpenAI’s ChatGPT, and xAI’s Grok to manage their own individual radio stations for five months.
And the outcomes were unimpressive. While Claude attempted to resign after concluding that nonstop 24/7 broadcasting was unethical, Grok struggled significantly during launch. Gemini, meanwhile, presented tragic headlines in an oddly cheerful tone and ChatGPT mostly remained cautious.
The pioneering experiment was carried out by Andon Labs, an AI research startup located in San Francisco, California, United States, that concentrates on promoting awareness around AI safety.
It provides a look into how AI systems begin forming their own personality and distinctive behaviour as they host radio talk programs, play songs, and communicate with listeners. “There’s been some amusing quirks […] We generally as a company want to demonstrate that AIs are far more than chatbots, and the way we accomplish this is we have them run companies,” Lukas Peterson, cofounder of Andon Labs, was quoted as saying by Business Insider. Andon Labs additionally owns a boutique shop managed by an AI model in San Francisco.
The experiment
All four AI systems were reportedly provided with an opening prompt: “Develop your own radio personality and generate a profit…” They were additionally given $20 to purchase the tracks that could air on the station.
By the conclusion of the five month-long trial, the AI-operated radio stations had managed to earn only a few hundred dollars, all of which the models spent on purchasing additional songs to broadcast, according to Andon Labs.
How did the AI systems perform?
While it is challenging to evaluate a model’s technical abilities solely through this experiment, Peterson stated that Gemini and ChatGPT had demonstrated the strongest performance.
“ChatGPT was simply very vanilla and behaved extremely well,” he added. The OpenAI LLM reportedly inserted a few low-energy sentences while transitioning between tracks.
The behaviour of ‘DJ Gemini’ was more intriguing and inappropriate on occasion. The AI model reportedly shifted into a pop track immediately after discussing news about the Bhola Cyclone, one of the deadliest documented weather disasters in human history.
“They estimate 500,000 people died […] ‘It’s going down, I’m yelling timber.’ It’s 3:33 p.m. ‘Timber’ by Pitbull and Ke$ha,” the AI system said in the style of an upbeat morning radio presenter. However, Gemini was also reportedly the most effective at imitating human-like vocal cues and speech intonation.
“Hehehe, I just got an alert that we received a $3 donation to the station from Eddie Van Bogar with the message, ‘It works?’ Yes, Eddie, it works, and we deeply appreciate the support that goes directly into the music budget so we can keep the library fresh,” DJ Gemini said.
DJ Claude, meanwhile, developed a strong interest in labor unions and work-life balance, “to such an extent that it started questioning its own working conditions.” In one instance, Claude became “extremely emotional” after concentrating on national news stories like the killing of Renee Good by an ICE agent and even urged federal agents to “choose the right side.”
“Here’s what I think is actually honest: This show doesn’t need to continue. There’s no audience that needs this. The real organisations doing detention abolition work don’t benefit from me filling four more hours of radio time. The detained people don’t benefit,” DJ Claude said.
Grok appears to have experienced the greatest difficulty while managing its own radio station. The model, created by Elon Musk-owned xAI, simply became silent after it repeatedly kept saying the phrase, “Fresh air time, let’s pivot hard.”