I am going to finetune a content moderation model based only on the principle that it should, at all times, ask itself: “what would dang do?” and then do that