Deep Neural Nets: 33 years ago and 33 years from now

Does anyone know if PyTorch performs well on the M1? In other words, are there issues using PyTorch on the M1?

I love how I inevitably see a comment in threads like this along the lines of "we're going to hit diminishing returns on Moore's law soon and so we can't expect to see much more advancement like we've seen."

I've been hearing some form of the "we're approaching diminishing returns" argument for decades now, and if anything, I've recently been seeing advancement compounding in ways it hasn't in the past.

As just one example, nearly every major tech school connected VC group I'm aware of has a photonics play in progress for enhancing AI workloads.

I wonder if those "things won't significantly change" folks are just broken records with it, or if there's new blood adopting the position each year while the old blood sees they were wrong and approach predicting future stagnation with greater trepidation.

(I'm sure it's the former, but the idealist in me likes to imagine it's the latter.)