Delving into Out-of-Distribution Detection with Vision-Language Representations Yifei Ming