None of our known theories of physics makes any predictions before specifying a model in the theory.
Think about it: There are many, many choices made to pick the standard model of particle physics out of the huge space of possible local QFTs. Think about how vast that space is, the "landscape of QFT": you can write down pretty much any local Lagrangian that is quantum anomaly free and get a QFT. Nothing in general QFT "predicts" that the world is described by Yang-Mills theory coupled to fermions with the gauge group and the coupling constants seen in the standard model of particle physics. These are a huge number of choices we humans make to fit QFT parameters. Only once all these choices are made, does the standard model start to predict anything.
That why it's called the standard model and not the standard theory! The standard theory is QFT, but that alone predicts nothing.
Same for the standard model of cosmology. Nothing in Einstein gravity predicts that the universe is modeled on large scales by an FRW model with positive cosmological constant. The back and forth with the assumptions made here is legendary (Einstein's "biggest blunder" in model building is now our Nobel worthy insight. Nothing predicted it.) Only after the FRW model is chosen, only after dark matter content is adjusted to fit observations, only then does Einstein gravity start to make predictions in cosmology. The "landscape of Einstein gravity" is huge and vast: think about all the possible solutions to Einstein's equations. There are not just 10^500 of them. It's certainly not a finite set, it's a hugely infinite-dimensional space.
One needs to remember this before getting worked up about the alleged lack of predictivity of string theory.
For in string theory it's the same: of course the theory itself doesn't predict anything. Just as QFT and Einstein gravity by themselves predict nothing. In each case one first needs to fix a model and then the theory make predictions about the remaining parameters.
The irony of this is: the constraints on model building in string theory are much stronger than in QFT. Not every model in QFT lifts to a model in string theory (the "Swampland" doesn't sit in the "Landscape"). On the other hand, string theory models generically contain Einstein-Yang-Mills theory. At least that. In QFT you can write down weird theories that have nothing to do with the real world. Like phi^4 theory. In string theory you cannot, since its much more constrained. String theory predicts that the world is not described by phi^4 theory. QFT cannot predict that. For instance.
So string theory is clearly more predictive than QFT! Think about it. In string theory we have a chance of arguing that there may be a finite number of models -- even if it is a large finite number such as 10^500. This is nothing against the hugely vastly infinite-dimensional space of choices of writing down local QFTs.
String theory, just as QFT and just as Einstein gravity, starts making predictions as soon as you specify a model. And that's precisely what string phenomenologist's do. They build models (or approximations to them, because in string theory its harder, since it is more constrained) and check against the known observational constraints. Then they refine their models. This process keeps going on check out the commented links at
http://ncatlab.org/nlab/show/string+phenomenology
Also check out the pertinent paragraphs at
http://ncatlab.org/nlab/show/string+theory+FAQ
And think about. Don't just repeat the "Oh my, oh my, string theory makes no predictions." that you hear in the blogosphere. Before you repeat this, think about what it means for physical theories to make predictions, how QFT makes predictions only once we choose the standard model.