What I really liked about this article is near the end, where it identifies two things: * Automatic differentiation of higher-order differentiation being impor

What I really liked about this article is near the end, where it identifies two things: * Automatic differentiation of higher-order differentiation being important, and how there's clearly room to disrupt there * Increasing hardware diversity seems to mean that both frameworks will run into a brick wall as-is Exciting space. It'll be fascinating to see how dramatically, or not, things change in the coming years.