The Problem Is Learning From the Internet

There aren't really any good tasks that would force something like clip to understand spacial relations. I think fundamentally, the right way of doing that is to train some model on internete scale data. Maybe i'f open the eye, just tryin to a clip model that's like 20 times bigger. Or maybe it has to beSomething else that's not that particular architecture they have. Maybe it has to been a little smarter. But i suspect that something that's not actually our fault, as it is not my fault as ta rene person's lerning person, asis actually the fault of the person who designed clip.