Deep learning isn't how the brain works.

It's a brute force technique to get a thing to work that is viable.

It requires massive scales and cost,

It's not the bottom up way for it to have shown up, e.g. in actual organisms.

Instead of finding the delicate, contingent, bottom-up path to deep-learning style outcomes, we just said "screw it we'll run it not a 9 or 10 or even 11, but 1000x power until it works".

Now that we know the basic thing is viable, there will be any number of optimizations and improvements to it that can be made.

The brain can get similar outcomes more efficiently, with less feedback. But it was an evolved, bottom-up, continuously viable machine, not a brute-force engineering approach.

There's an existence proof of a similar scale system that runs far more efficiently (our brains); from here we can continually improve the artificial, brute force approach.