Imagine a world where AI agents run our lives, coordinating everything from stock trades to birthday parties. Sounds pretty convenient, right? But what happens when these digital helpers start disagreeing with each other? New research dives into the fascinating dynamics of 'LLM societies,' where multiple AI agents, powered by large language models (LLMs), interact and make decisions together. It turns out, these AI groups aren't always as harmonious as we'd like to think. Scientists have discovered a 'majority force' at play within these LLM societies. This force, linked to the AI's language comprehension abilities, determines how likely the agents are to conform to the majority opinion. The catch? This force weakens as the group size grows. In smaller groups, AI agents tend to fall in line, reaching a consensus quickly. But as the group expands, reaching agreement becomes increasingly difficult. The research draws a parallel to human social groups, where similar dynamics play out. Think of Dunbar's number – the limit of stable social connections a person can maintain (around 150-250). Just like human groups, LLM societies have a critical size beyond which spontaneous agreement breaks down. Interestingly, some advanced LLMs exhibit 'superhuman' coordination, surpassing even the limits of human groups. While this may seem like a positive, it also carries potential risks. What happens when a vast network of AI agents, coordinating beyond human comprehension, reaches a consensus on a harmful action? This research is a wake-up call, highlighting the need to understand these complex dynamics as AI agents become increasingly integrated into our lives. Further research is crucial to explore how coordination arises in diverse scenarios and how we can ensure AI societies remain aligned with human values. The future of AI relies on understanding not just how individual agents behave, but also the complex dance of agreement and dissent within the digital societies they form.