Also, in case it helps anybody else understand what "Faster Training = Better AI" is getting at:
I initially thought it was implying that there were AI training frameworks that ended up with more accurate classification metrics after briefer periods of training. That's not it!

In 2018, it took 6.2 minutes to train the best system; in 2020 it took 47 seconds. This extraordinary improvement was enabled by the adoption of accelerator chips that are specifically designed for machine learning.

The point being made is actually that faster training reduces the risk of applying research efforts in any particular direction, which inherently streamlines the idea-to-impact pipeline for researchers. Presumably, the relatively lower cost provided by faster training mechanisms is also helpful.