Werewolf Game

The Werewolf game, a social deduction game requiring deception and deduction, serves as a compelling testbed for evaluating large language models (LLMs) in complex, interactive environments. Current research focuses on developing LLM-based agents that can effectively engage in strategic communication, employing techniques like rule-based systems combined with LLMs, reinforcement learning to optimize decision-making, and weighted ensemble learning to aggregate diverse strategies. These studies aim to improve LLMs' abilities in natural language understanding, strategic reasoning, and persuasion, with implications for advancing AI in areas like human-computer interaction and multi-agent systems.