From 71822faf431963e3f4bee51dd08e1e6667f06cb7 Mon Sep 17 00:00:00 2001 From: mmorge <maxime.morge@univ-lyon1.fr> Date: Thu, 17 Apr 2025 17:35:11 +0200 Subject: [PATCH] LLM4AAMAS: Add hua24arxiv --- README.md | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/README.md b/README.md index 3e6bf8a..7a257d6 100644 --- a/README.md +++ b/README.md @@ -563,6 +563,21 @@ still show limitations in adaptability. Guan, Tao Ge, Furu Wei (2025) at *COLING 2025*. + +This analysis shows that LLMs often stray from rational strategies, especially +as game complexity increases. Notably, the nature of their irrationality differs +from that of humans. The study highlights a key limitation: LLMs struggle with +complex strategic reasoning unless provided with explicit structural support. To +address this, the authors propose game-theory-inspired workflows that guide +LLMs’ reasoning using principles such as dominant strategies, backward +induction, and Bayesian belief updating. Interestingly, the model that initially +performed the worst without the workflow becomes the best with it — and vice +versa. + +- **[Game-theoretic LLM: Agent Workflow for Negotiation Games](https://arxiv.org/abs/2411.05990)** + Wenyue Hua, Ollie Liu, Lingyao Li, Alfonso Amayuelas, Julie Chen, Lucas Jiang, Mingyu Jin, Lizhou Fan, + Fei Sun, William Wang, Xintong Wang, Yongfeng Zhang (2024) on *arXiv*. + ### Generative MAS on the shelf - [MetaGPT](https://github.com/geekan/MetaGPT) is a framework for creating -- GitLab