As someone who spends 98% of his time in Python and the remaining 2% in R I'd love to switch to a more nicely designed, ergonomic, high-performance language. Adopting Julia at work is a nonstarter for now, but what really holds me back from learning it is how thoroughly academic all the evangelism seems to be. Most working data scientists aren't doing things like physics simulations and won't be terribly interested in solving differential equations; even in grad school the topic rarely came up for my econometrics research.
To put it bluntly: reading about Julia kind of makes me feel a bit dumb, not excited.
When I do work with neural networks, the networks just aren't going to be small because the value in them -- for the kind of work I do -- is in their capacity to use large amounts of data to find useful representations. So the benefit of Julia over PyTorch described in the link is only interesting to me in a "huh, I guess that's a bit cool" kind of way.