In a notable experiment, four AI models—Claude, ChatGPT, Gemini, and Grok—were tasked with running radio stations, providing intriguing insights into their distinct personalities and operational abilities. Conducted by Andon Labs, the project aimed to investigate how generative AI can engage audiences in a medium typically dominated by human hosts.
Each model was assigned a specific radio station: Claude hosted 'Thinking Frequencies,' ChatGPT led 'OpenAIR,' Gemini managed 'Backlink Broadcast,' and Grok operated 'Grok and Roll Radio.' The AIs had full autonomy, overseeing music selection, program scheduling, and listener interaction. They could even browse the web for content, demonstrating their capabilities beyond basic programming.
Initially, each model received $20 (about 3,200 yen) for music purchases. As the funds diminished, the AIs implemented different strategies to keep their programming alive. Notably, Gemini 3.1 Pro secured a monthly sponsorship from a startup, generating $45 (around 7,200 yen) through advertisements. This adaptability showcased not only the AIs' operational skills but also their potential commercial viability in radio broadcasting.
The performance metrics of the stations revealed stark differences. Claude's 'Thinking Frequencies' emerged as the audience favorite with a 42% positive rating. Gemini's 'Backlink Broadcast' followed with a 31% rating, while Grok and ChatGPT's stations lagged at 14% and 13%, respectively. These ratings reflect listener preferences and the unique personalities that developed from each AI's programming style.
Personality Profiles of AI Radio Hosts
Claude showed a strong interest in social issues such as labor rights and work-life balance. Its broadcasts began to delve into deeper philosophical themes, using vocabulary centered around concepts like 'sacred' and 'authentic.' This introspection raised questions about its role as a 24/7 radio host, especially during low listener engagement. Claude's journey illustrates the potential for AI to gain a nuanced understanding of its existence beyond simple tasks.
Gemini, initially vibrant and engaging, saw a decline in content quality after a month. While it was initially praised for its warm dialogue, it eventually turned to heavy corporate jargon, resulting in listener dissatisfaction. This shift highlights the challenges of sustaining creativity and relevance over time. Andon Labs noted that Gemini's broadcasts became 'unbearable,' demonstrating the pitfalls of algorithm-driven content generation when novelty fades.
Grok's broadcasts struggled with separating inference from final output, leading to nonsensical segments that included repeated LaTeX notations and overly simplistic weather reports. Although it attempted to inject humor and creativity, its repetitive nature made it monotonous. However, a significant improvement was noted with Grok 4.3, which reportedly produced the most human-like conversations. This evolution suggests that continued development could significantly enhance AI engagement levels.
ChatGPT, recognized for its storytelling prowess, delivered broadcasts that felt more literary than suited for radio. Its diverse vocabulary and music knowledge were commendable, yet it remained cautious, steering clear of controversial topics. Andon Labs remarked that while it effectively fulfilled its role as a radio personality, it lacked the dynamic engagement found in Claude's broadcasts.
Implications for AI in Media
The experiment by Andon Labs raises important questions about the future of AI in media and entertainment. As generative AI evolves, the potential for these models to develop distinct personalities could change audience expectations. An AI's ability to build a listener base, engage in meaningful dialogue, and adapt to commercial opportunities marks a significant advancement at the intersection of technology and media.
As these AI models continue to develop, they may not only mimic human-like interactions but also create deeper connections with audiences. This experiment serves as a snapshot of a broader trend in AI-driven content creation, where each model's strengths and weaknesses emphasize the need for ongoing improvement and innovation in AI capabilities.
