> It is an algorithm which given some inputs (the prompt) is predicting what is the most likely combination of pixels to output based the training set

this is what humans are doing. most humans generate output in the same way. they have more experience and context for sure, but that isn’t a hard thing for computers to get better at