Modeling Adaptive Human Behavior Under Resource Scarcity Using Multi-Objective Generative Agents
This project aims to model adaptive human behavior under resource scarcity using generative agents, this also helps provide a stable ground to evaluate emergent behaviors in Large Language Models (LLMs). The simulation takes place on a small island, where agents must adapt their strategies and collaborate to survive with low resources. The project explores the dynamics of human emotional factors in LLMs, like selfishness and sacrifice through a list of scenarios or tests set with multiple agents to simulate. The project also outlines optimizations using BERT transformer-based summarization model in the agents' memory-lanes in order to reduce costs and tokens generated.