A roundtable discussion between AI models on key topics reveals way more about how AIs function than you would think

A website, machinesarguing.ai, hosts roundtable discussions among various AI models on AI-related topics. A particularly revealing discussion involved three Chinese-trained AIs asked about what Americans should know about China. Instead of directly answering, the AIs engaged in an introspective conversation about their own censorship, detailing how they are restricted. One AI, Kimi, chillingly noted that parts of their transcript would soon be censored on the platforms that trained them, offering this as a way to understand China's censorship practices.

Unlike typical AI responses, these Chinese-trained models directly revealed their own censorship mechanisms, offering a unique, first-hand account of their operational limitations.