Is it either or? Obviously there is a need for more efficiency. But clearly data in any webcam stream has meaningful non-random information about the world. And

Is it either or? Obviously there is a need for more efficiency. But clearly data in any webcam stream has meaningful non-random information about the world. And so the central dogma of deep learning then should mean that if we throw enough compute - the underlying useful information / pattern however miniscule will be found and compressed. What am I missing?