Can an AI model predict perfectly and still have a terrible world model?