As someone working in "scientific machine learning", I've found that the Python-vs-Julia debate comes up quite a lot!
This is an interesting recent thread on the state of ML in Julia, in particular with reference to PyTorch, JAX etc.:
@PatrickKidger
It was a nice read! I would add 2 things:
1. For CPU even Numpy can be faster than vanilla Julia and its simpler to write, fast Julia requires macros and becomes less readable. This is not good for day to day stuff. JAX CPU has an amazing performance vs simplicity tradeoff.
@PatrickKidger
One point to add here (which might be obvious to some), the core point stands that there is a massive amount of momentum behind JAX & PyTorch, and I do find it an open, but yet intriguing question, whether new projects such as Dex will take momentum out of the Julia ecosystem...