Earlier quoted context omitted.
I argue that what makes the sale, is that that is what so many in the field are using. Not much inherent to the technologies, but you are taking on a daunting task if you want to compete with pytorch, instead of using it.
But we can assume people got on python/numpy stack because it brought some value compared to R. I'm not a big fan of numpy but I find it a bit shallow to think it's mostly follower effect at play. It might be though.. but that would be sad. Maybe Julia will bring balance back in the universe.
Well, R is a rather awkward programming environment. And many folks learned on Mathematica and Matlab.
What python brought to the table, was that it was freely installed on most machines.