AI Populates Virtual Town and Behaves Like Believable Humans

Researchers from Stanford and Google created a virtual town populated with 25 instances of ChatGPT, a machine learning model, to produce "believable simulacra of human behavior." The agents were prompted with information about their circumstances and given prompts to simulate realistic actions. The experiment was successful in creating a virtual environment where the agents interacted with each other in a believable way. The implications for simulations of human interactions are huge, but the approach is still impractical for games and virtual environments.