Evaluating Creativity and Deception in Large Language Models: A Simulation Framework for Multi-Agent Balderdash