I've watched the Harvard's sneak preview of WolframAlpha if it does about 1/3 of what the presentation actually shows (it's very easy to tailor a presentation to things you know your software can do and hide things it can't), it would be impressive. However, there are some odd things that I'm not sure make much sense. One of the reasons I can't be sure is that the presentation itself was terribly recorded. You can basically hear Wolfram talking and typing on a computer, saying abstractly what he is receiving back, but they don't really show the images.
But in some cases he gives some fishy examples. For instance (as we are talking about fishy things), his question of "How much fish is produced in France?". It starts with useful information and then he says that it's "1/5th of the rate that trash is produced in New York City". Pretty, but why would you waste computing power to show such randomly useless data? It doesn't seem right, unless there is something else that is not being shown, or if it's not doing what it should be doing.
In general I'm eagerly waiting for it. Just thinking that I'll have a free Mathematica to play with, it's already worth it. I'm sure this they can get right! And if it can calculate the number of people that travel by plane from Oklahoma City every day divided by the square root of the average number of years that a French monarch stayed in power between 1224 and 1843, all the better! I always wanted to know that!