it has to do with world model perception. these models don't have it but some can approximate it better than others.