1. It makes writing UI state machines much simpler since it is declarative. Instead of saying "when this information changes, update the UI", you just say "This bit of UI shows this information" and then it updates it does it automatically.
2. Animation is also much easier (and possible at all).
3. Layout is... myeh it is not so good actually. UI files made it really really simple, but QML has a kind of Android-like overcomplication to it. For example I have no idea how to have two buttons above each other with the same width and have that width the minimum required to fit them both. With the old system I didn't even have to think about that.
4. Javascript was a seriously bad choice. Fortunately you don't write that much of it (and the non-javascript parts of QML are well designed). The biggest problem is that you have to run code to have even the vaguest idea that it will work (probably you misspelt a variable). With C++ if it compiles you at least have some idea that it will work. Also there is a lot of hassle converting between C++ values and javascript objects. Particularly for lists/vectors of things which work very badly and... to be honest does not really work.
5. It does force you to separate your logic and UI - kind of like HTML/CSS. This is a good thing generally.
6. Once you've done the hard C++ work, building the UI is much quicker and less verbose than using the old system. It is also easier to change around.
In generally they are going in the right direction. They just made a huge mistake by choosing javascript. They've ended up writing their own javascript VM too, so I don't know why they didn't take the opportunity to use a sane, strongly typed language.