1 paper
Wei Shi, Ziheng Peng, Sihang Li +4
LLM agents exhibit a consistent tendency to over-call, invoking tools even in situations where none is needed. On the When2Call benchmark, six models from three families show high…