Give 1,000 chatbots a debate and they'll sort themselves into echo chambers all by themselves

What happened

Researchers at Tsinghua University simulated networks of 1,000 LLM agents, mainly GPT-3.5, discussing party alignment, gun control and abortion bans. Over 40 rounds, agents increasingly interacted with like-minded peers and opinions polarized, and some agents broke character. Interventions such as receptivity prompts reduced the polarization score by about 12%, but the authors say the results are hypothesis-generating.