UNO Arena for Evaluating Sequential Decision-Making Capability of Large Language Models

Sequential decision-making refers to algorithms that take into account the dynamics of the environment, where early decisions affect subsequent decisions. With large language models (LLMs) demonstrating powerful capabilities among various tasks, we cannot help but ask: Can Current LLMs Make Sequential Decisions Effectively? In order to answer this question, we propose the UNO Arena based on the card game UNO for evaluating the sequential decision-making capability of LLMs and explain in detail why we choose the UNO game. In the UNO Arena, we also involve some novel metrics based on Monte Carlo methods for evaluating the sequential decision-making capability of LLMs dynamically. Besides, we set up random players, DQN-based reinforcement learning players, and LLM players (e.g. GPT-4, Gemini-pro) for comparison testing. Furthermore, in order to improve the sequential decision-making capability of LLMs, we propose the TUTRI player, which can involve enabling LLMs to reflect on their actions with the summary of game history and the game strategy. Various experimental results demonstrate that the TUTRI player can achieve a notable breakthrough in the performance of sequential decision-making compared to the vanilla LLM player. 1