I like it. The fact that f(x)=x^2 without further ado defines a function that works properly for several different types of the argument x is very nice. And automatic differentiation is simple and intuitive to use. Caveat - I'm a lightweight in this domain.
Traditionally, maths applications were written in Fortran and in the 90s, C++. The object-oriented paradigm reigned but the match with maths was less than optimal. Later, attempts were made to apply templates, but it is the wrong paradigm. Very awkward. The 3rd paradigm Functional Programming (and lambda calculus) was developed to reduce this cognitive gap.
Julia seems to tick all the boxes in this regard (altho' it does not seem to be quite FP). I used to work in oil and gas,One of my clients/students is applying PDE to reservoir simulation an we will probably use Julia as the programming. Many methods from computational can be used here as well. in fact, the well-known ADI was invented by Douglas, Jr. , Peaceman and Rachford who worked in the oil industry.
Exsan: Haskell is a quiche^2 language IMHO. But is a real FP language.