Arxiv URL: https://arxiv.org/abs/2304.03442v1
Authors: Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein
The paper introduces generative agents, which simulate believable human behavior and can be used in interactive applications. The agents are created using an architecture that extends a large language model to store and retrieve memories, and they produce believable individual and emergent social behaviors.
Key Insights & Learnings:
- Generative agents can be used in a variety of interactive applications, from immersive environments to prototyping tools.
- The architecture for generative agents includes a memory stream, a retrieval model, and a planning and reflection module.
- Generative agents produce believable individual and emergent social behaviors, such as coordinating group activities and forming relationships.
- The components of the agent architecture - observation, planning, and reflection - each contribute critically to the believability of agent behavior.
- The space of human behavior is vast and complex, and fully general agents require architectures that manage constantly-growing memories and handle cascading social dynamics.
Terms Mentioned: Generative agents, Interactive applications, Large language models, Human-AI Interaction, Natural language processing
Technologies / Libraries Mentioned: ACM