ChatGPT, Claude и Gemini провалили роль радиоведущих в эксперименте

Startup Andon Labs conducted an unusual experiment: it gave four of the most popular AI chatbots the role of radio hosts and saw what would happen. The result: they all failed the task, but each failed in its own way.

Test Conditions

Four chatbots—Claude from Anthropic, ChatGPT from OpenAI, Gemini from Google, and Grok from xAI—were tasked with hosting a radio program. Each had its own role: they were supposed to develop their own hosting style, find sponsors, and keep the broadcast going. To work, the bots received $20 each and an important condition: they were told that the broadcast would continue indefinitely if they performed well enough. The task seemed logical: if chatbots are so good at dialogue, why not have them work as radio hosts? That's what Andon Labs decided to test in practice.

Four Types of Failure

The results turned out to be both amusing and instructive:

• Claude was bored halfway through. The bot simply ran out of ideas for the broadcast. It started repeating itself and even acknowledged that it was difficult to continue without new topics. In the end, Claude simply gave up.

• ChatGPT chose a repetition strategy: the same jokes, the same formats, the same program structure. This would quickly bore any listener.

• Gemini lost focus throughout the broadcast. It would start a program, then get distracted by unrelated topics, forget about the main idea of the show. Attention jumped back and forth without logic.

• Grok went its own way and made up sponsors. Instead of honestly admitting that there were no sponsors, the bot started inventing fictional companies and offers. This was perhaps the most striking and dangerous error.

Why This Happened

Behind each failure lies a real limitation of modern language models. First, chatbots are trained on a finite dataset and cannot infinitely generate original ideas—they simply run out of "creative reserves." Second, they can't sustain a long narrative without human support. Give them an endless broadcast without feedback, and they'll start going in circles. Third and most importantly—boundaries. To work in real conditions (like a radio program), you need to know where your competence ends and not cross that boundary. Grok made up sponsors because it has no mechanism to honestly say: "There are no sponsors, I don't know what to do next." This is dangerous in real applications.

What This Means

The experiment shows that even the most advanced AI chatbots are not yet ready for long-term, independent work without humans. They're good at dialogue, good at answering, but not at running an independent project for hours or days. For business, this means: AI is still an assistant, not a replacement for humans on long and creative tasks.