It does. In general this is known as teacher-student training or knowledge distillation. It works better if you have access to the activations of the model but you can work with just outputs as well.