1 paper
Yao Liu, Yu He, Jihao Huang +2
Playing the user with a large language model is standard in scalable evaluation, and has a repeatedly diagnosed failure: the simulated user is excessively cooperative, so a system…