(There is also the original benefit discussed in #8450 <https://github.com/JuliaLang/julia/issues/8450>: when you write a function f(x::Number), you don't have to decide in advance whether you need a separate "vector" version of it too. You can just write f.(x) for any scalar function f.)
Note that all of this is not really an option in Python or NumPy, because Python loops are too slow — for performance, they must decide in *advance* which loops they want to vectorize, and implement those in C/Cython/etc. There are also various code-generation/JIT frameworks for NumPy that can fuse NumPy loops, but again they can only fuse a set of NumPy operations that are selected in advance, not arbitrary user-defined functions. In consequence, in Python there is not as much practical benefit to be gained from a syntactic broadcast.
