There was a viral video from Alex O Connor, its old at this point (for our modern era a year is a lifetime) where he tries to get a frontier model, make an image of a cup of wine filled to the top. In other words, about to spill.
No matter the prompt, no matter the guardrails it failed hilariously every single time. Why? because it simply had "never seen" a cup of wine filled all the way, and thus could not infer it.
I can find it if you are curious. It also made me laugh a bit, like your ontological battle here, which is why I remembered.
RE: The Racket Problem