CollabBench benchmark measures LLM collaboration with diverse players

Arxiv CS.CL (NLP/language models)

CollabBench is a new benchmark evaluating LLM agents' collaborative ability through grounded interactions with simulated human partners. It includes diverse player types and requires proactive engagement beyond simple conversational collaboration.

llmcollaborationbenchmarks

Arxiv CS.CL (NLP/language models)·Hong QianYuanhao LiuZihan ZhouZongbao ZhangHanjie GeHaotian ShiLiang DouXiangfeng WangJingwen YangAimin Zhou

6 days ago