This uses a lot of concepts like "common sense" and "understanding" that I'm not sure are useful when talking about language models.