Feels like LLMs could be used to scalably grade more complicated assignments than multiple choice tests.