I've been reading about Wolfram|Alpha on this article on TechCrunch. The article's author is certainly a little too excited about it (Nova Spivack, the guy behind twine), but I think there is something to his excitement. Most commenters don't get it at all. Every time they hear something about somebody building a product to answer questions on the web, they think that it's a bust. History taught us that it's very likely to be a bust. Why would Wolfram have it different?
This is not the blog to go in details about this, but I think that, like most projects out there, they will only learn by trying. And it's not like other startups that build something like this hoping that they would be bought before they have to actually make it useful, Wolfram is probably a little too big to be going this path. Also, they have a name. Not a huge name outside the scientific community, but it's a brand anyway. So, they need to be careful with what they release. (what a terrible reason to be hopeful about a product, I know...)
But back to a high-level personal opinion: I hope they show us something new and not just flashy answers to the questions they've trained on and nothing else (that people like me can find). I've seen too much of this on [true knowledge] and it drives me crazy. I'll eventually go back to my project on modeling knowledge. I've been away from it for way too long.