“A raw, pre-trained-only language model is difficult to use and appears less capable compared to a model that has undergone post-training.”