1 paper
Weixiang Sun, Zehong Wang, Hong Huang +3
Multi-agent systems built from large language models are deployed widely, yet how much performance is lost when two LLMs must coordinate rather than act alone remains unclear. We f…