"frees the programmer from writing superfluous type signatures" is a weak (and dubious) advantage. I very often write "superfluous" type signatures first (to be sure I know what I'm asking my program to do) and only then let Haskell check it. Then I leave it in as good documentation.
I agree with this. Perhaps
Type Inference: deduces types automatically, so you don't have to clutter up your code with type declarations. You can still write type declarations for documentation purposes, and these will be automatically checked by the compiler.
I also write top-level signatures usually, but inference is still really nice inside the function. Things like "let x = f y in ... + g x + ..." don't make you declare the type of 'x' and change the declaration if 'f' changes return type. You just have to change 'g', not everyone in between 'f' and 'g'.