Presumably you mean "better", not "worse". Well, it's better because you will (or should) get a compiler warning (or ideally an error) about the latter version.
I didn't mean to imply that dynamic languages are "better" because they allow unchecked combinations of everything with everything else. Although I also didn't mean to imply that they're strictly "worse", precisely because there are things we can express easily in dynamic languages which take effort to reproduce in something like Haskell.
My main reason to mention dynamic languages was as motivation for why you might want to write such code in the first place: using dependent types to, say, prove Euclid's theorem of the infinitude of primes, would probably not motivate many software developers to give Idris a try. Examples like the type-safe printf mentioned in a sibling comment would presumably be much better motivators: we compute a type based on the format string, so providing a string like "Hello world %f %s !" will produce a type `Float -> String -> String` which accepts the required Float and String parameters, then returns the resulting String.
Off-topic, but related to undefined in Haskell, I love holes in Idris as a way to represent incomplete programs. I'd love if it somebody could take them one step further and have it so that if a hole is evaluated at runtime, a prompt came up and said "Well this is the input, what do you think the output should be?". You type it in, the program keeps going and a unit test is generated and put away somewhere for you to make sure that later when you implement it, you can help verify that you're doing it right.
For some cases, this is pretty simple to implement with unsafePerformIO - I've done that once or twice. Not a sin at all in development.
This makes sense on its own, but seems a bit silly when we consider that completely removing the signature will fix the warning ;)
As somebody with familiarity with Haskell in the small but not in large applications, is it not similarly trivial to write in Haskell with the same level of safety that dynamic languages give you (i.e. explosion at runtime)?
Consider as an example the following safe lookup function (standard Haskell)
and the following unsafe lookup function How is the latter any worse than lookup in a dynamically-typed language where you are assuming the lookup will succeed (and hence don't write any error handling logic)? I know this sort of code is frowned upon in Haskell, but it can be done.