Lesson one: nothing in which the text matters, as dall-e 2 clearly has no language capability at all. That the model didn't learn that words in the prompt are best transcribed is perhaps not entirely surprising, but it sure is glaring.