Ikka,

> In about the next five years, we use roughly the same amount of
> energy for deep learning than for cement production, globally.
> This, in my view, is a very concrete indicator...

Was that an intentional word play? Either way, it was good.

More seriously, this assumes that deep learning will continue to
dominate the field. Applications of machine learning have relied
on far more efficient methods in the past, and it seems very likely
we'll develop more efficient ones in the near future.

Regards, -Pat