Can an AI model predict perfectly and still have a terrible world model? | Dark Hacker News