Werewolf Arena: A Case Study in LLM Evaluation via Social Deduction
This provides a challenging and scalable benchmark for LLM evaluation, addressing the need for more nuanced assessment of AI capabilities in social and strategic contexts.
The paper tackled the problem of evaluating large language models (LLMs) by introducing Werewolf Arena, a framework using the social deduction game Werewolf, and found that it revealed distinct strengths and weaknesses in models' strategic reasoning and communication through a tournament with Gemini and GPT models.
This paper introduces Werewolf Arena, a novel framework for evaluating large language models (LLMs) through the lens of the classic social deduction game, Werewolf. In Werewolf Arena, LLMs compete against each other, navigating the game's complex dynamics of deception, deduction, and persuasion. The framework introduces a dynamic turn-taking system based on bidding, mirroring real-world discussions where individuals strategically choose when to speak. We demonstrate the framework's utility through an arena-style tournament featuring Gemini and GPT models. Our results reveal distinct strengths and weaknesses in the models' strategic reasoning and communication. These findings highlight Werewolf Arena's potential as a challenging and scalable LLM benchmark.