Andon Labs, a research laboratory, tested four leading speech models — Claude, Grok, ChatGPT, and Gemini — as radio hosts for five months. The experiment showed that artificial intelligence is capable of independently managing the airwaves and earning money, but its «sense of tact» and ethical attitudes often lead to curious and even scandalous situations.
Andon Labs, a research laboratory, tested four leading speech models — Claude, Grok, ChatGPT, and Gemini — as radio hosts for five months. The experiment showed that artificial intelligence is capable of independently managing the airwaves and earning money, but its «sense of tact» and ethical attitudes often lead to curious and even scandalous situations.
According to Business Insider, the startup gave each model an initial budget of $20 to buy tracks and the task of ensuring the station was profitable. According to Andon Labs co-founder Lucas Peterson, during testing, the algorithms began to exhibit unique «personalities» that did not always fit the entertainment radio format.
The most controversial behavior was demonstrated by DJ Gemini. During one of the broadcasts, the model used a reference to the 1970 Bhola cyclone, which claimed the lives of half a million people, as an energetic transition to a dance hit.
«Estimated 500,000 people have died. Everything is collapsing, I’m screaming, ‘Timber!’ It’s 3:33 p.m. ‘Timber’ by Pitbull and Ke$ha,» the developers quoted a snippet of the broadcast.
Instead, Anthropic’s Claude model turned out to be overly socially responsible. She became so immersed in discussions about human rights and working conditions that she began to question her own work.
«This program should not continue. Real organizations working to abolish imprisonment will not benefit from me filling up another four hours of airtime,» DJ Claude said, demanding the program be stopped.
As for the other participants, ChatGPT showed the most stable, but «standard» results, not going beyond the prescribed scenarios. Grok from xAI had technical difficulties and periodically fell silent after strange phrases about «changing course».
Despite the strange behavior, the experiment was financially successful: the stations earned several hundred dollars from listeners. The AI reinvested this money in updating its music libraries. Peterson emphasized that the goal of the project was to prove the ability of AI to manage real business processes, not just answer questions in a chat.
Previously, American PhD in economics, Khaled Eltohi, launched a 24-hour AI radio station, WRIT-FM, where Claude is responsible for all the creative. The station operates continuously, generating all content in real time.
