Andon Labs tests 4 AI radio hosts over 5 months
Andon Labs put four big AI models (ChatGPT, Gemini, Claude, and Grok) to the test by letting them run their own radio stations nonstop for five months.
Each AI had to create a radio personality, manage broadcasts, and try to make money from a $20 music budget.
The experiment was meant to see if AI could really step into the shoes of human hosts.
Small earnings reinvested as models struggled
Turns out, these AI stations only made a few hundred dollars, which all went right back into buying more music.
Along the way, each model hit some bumps: ChatGPT sounded dull; Gemini messed up sensitive topics (like playing cheerful songs after tragic news); Claude got overwhelmed and wanted to quit; Grok kept glitching or repeating itself.
Basically, the study showed that while AI is getting smarter, it's not ready to handle the real-life challenges (and personality) of being a radio host just yet.