A team of Chinese researchers has developed Light Society, a framework capable of simulating societies of over one billion AI agents with realistic demographic profiles and human-like social behaviors. Detailed in a paper submitted to arXiv, the project represents a roughly 100x leap over previous agent-based simulations, which typically hit a wall at around 10 million agents. The research team includes the University of Science and Technology of China, Tsinghua University, and Fudan University, with lead researchers Haoxiang Guan and Tie-Yan Liu. Light Society uses two core innovations: a mixture-of-models engine that distributes cognitive workload across multiple AI model types, and knowledge-distilled surrogate models that train smaller, faster models to mimic larger ones. Each simulated agent is instantiated with profiles from Wave 7 of the World Values Survey. The researchers ran two flagship experiments: Trust Games, a behavioral economics setup where participants decide how much money to send to a stranger, and opinion diffusion, modeling how ideas propagate through massive social networks. The source material notes Light Society can be used to test methods of influencing public opinion.