“Just adding multithreading alone is enough to wipe the floor with an otherwise comparable AI on a large machine. If you stack in 10-20 more optimizations that seem too annoying for most people to do by hand, the difference in strength is tremendous.”
Luu measures instead of asserting, which puts this ahead of most performance writing. The argument is that optimization got cheap, not that software got faster. Those are different claims and only the first one has evidence behind it. Nothing here explains why anyone would spend the newly cheap effort when shipping features still pays better.