We investigate the phenomenon of few-shot expert identification in large Mixture-of-Experts models and propose EASY-EP, a simple yet effective method for expert pruning.