A mildly interesting analysis of what models know and what it tells us about how they were trained.