Your Co-Workers Matter: Evaluating Collaborative Capabilities of Language Models in Blocks World

Abstract

Language agents that interact with the world on their own have great potential for automating digital tasks. While large language model (LLM) agents have made progress in understanding and executing tasks such as textual games and webpage control, many real-world tasks also require collaboration with humans or other LLMs in equal roles, which involves intent understanding, task coordination, and communication. To test LLM’s ability to collaborate, we design a blocks-world environment, where two agents, each having unique goals and skills, build a target structure together. To complete the goals, they can act in the world and communicate in natural language. Under this environment, we design increasingly challenging settings to evaluate different collaboration perspectives, from independent to more complex, dependent tasks. We further adopt chain-of-thought prompts that include intermediate reasoning steps to model the partner’s state and identify and correct execution errors. Both human-machine and machine-machine experiments show that LLM agents have strong grounding capacities, and our approach significantly improves the evaluation metric.

Publication
The 62nd Annual Meeting of the Association for Computational Linguistics (Findings)
Guande Wu (吴冠德)
Guande Wu (吴冠德)
CS Ph.D. Student

Hi! This is Guande Wu, a Ph.D. student in Tandon School of Engineering, New York University. My advisor is Prof. Claudio T. Silva and I am also working with Prof. Chen Zhao. My research interest mainly lies in the human-AI collaboration especially in AR scenario. Previously, I have worked with many outstanding experts in visualization and software engineering at Zhejiang University, Tongji University, UC Davis and Microsoft Research Asia and Adobe Research.

Related