An LCM actually understands concepts rather than simply having the ability to explain them or to draw an image.  This greatly reduces what I call "semantic aliasing" - the thing people often call "hallucinations."

An LCM provides the machine with world models, so it truly knows things like the effect of gravity, wind resistance or even things like a deer beside the road is likely to jump into the road and will not necessarily stay put or to go the other way.

This kind of understanding will help the system render realistic images and videos.