In some sense, in MWI there's no such thing as prediction, because anything is possible as long as it does not violate some conservation law.OK, that's more or less what I figured, and in your saying a prediction has to be probabilistic I would say, then, it is no longer really a prediction, merely an estimate of probability, which is a lot softer. A semantic quibble.
But the nice thing about quantum mechanics is that, even though anything is possible, the vast majority of outcomes are incredibly improbable. The probable outcomes are the ones that appear classical on the face of it (an egg falling down on the floor and breaking - classical & likely, an egg falling through the floor - possible but very unlikely).
Sometimes quantum events can lead to macroscopic bifurcations of paths and then you get multiple likely outcomes. There's even an iphone app for that: