Not surprised, I had low priors on the "magic" of certain ML architectures to begin with, and it's great to see experimental evidence supporting that skepticism. https://www.reddit.com/user/bmarti644