But fundamentally Idris is theoretically more advanced than Haskell. The core differences are
1. Idris functions can be proven to terminate (if you choose).
2. In Idris, types are first class values, and you can have dependent functions: functions whose return type depends on their input value.
An example of something you can do in Idris and not* in Haskell, in Idris you can define a vector type Vect n a, which is the type of vectors of length n with values in type a. You can also define Fin n, the set of integers less than n. Then you can define a function index : Fin n -> (Vect n a) -> a which takes an integer less than n, a vector of length n with element of type a, and returns an element of type a. This function is guaranteed to return a value, because the index is guaranteed to be in the correct range.
*For some meaning of "not": you can probably do this in some way in Haskell.
There are a lot of other differences between Haskell and Idris, though. Totality is a pretty big thing, and Idris is also strict by default. It also has support for proof tactics, which I don't see Haskell getting anytime soon.
> This means that whenever I call this function, I need to provide together with a and b a proof that b isn’t zero.
What might such a proof look like? And is this supposed to work at compile-time?
Accusing people of being "arrogant" is a cheap way feel righteously indignant at the expense of someone smarter than you. I was fortunate enough to have some very nice teachers in grades 11-12, who complimented me on my intelligence and didn't try to take me down a notch for the sake of their own egos (nothing wrong with the teachers below grade 11, these issues just didn't come up as much at that time).
I studied mathematics as an undergrad, and later got into programming. Now I do machine learning as my job, and study dependently typed programming languages for fun. If you like mathematics I highly recommend Haskell and Idris (or Coq or Agda, but I found Idris them most approachable, as a programmer). In 50 years I think everyone will be using something dependently typed languages (or some other kind of language that is also fundamentally different to existing languages).