9 months ago, in this article, I wrote about that many companies were trying to beat Google and detailed 3 questions towards this goal.

Today, I read about a new website coming in May, made by Stephen Wolfram (who made Mathematica), namely Wolfram Alpha.

Wolfram himself announced and introduced at this blog , and Nova Spivack, who made Twine, had chance to spend two hours with Wolfram being demoed the website, and then commented back about it.
Right after, Jon Stokes, who I cherish the damn insightful articles around chips at Arstechnica, expressed quite abruptely his doubts about the site.

Let's look at this site through my 3 questions:

1) Doing better
On this point, Wolfram Alpha (WA) seems to be on a right track. Following exactly the path I laid out (coincidence obviously ;) ), WA will be to give ANSWERS to requests, NOT just PAGES with the words told.
You will type questions in a typical Google-like search bar and WA will answer you.
You might so not cobble with the toundra of pages from Google to find the answer you're (implicitely) looking for when telling words to google.com.

2) Doing it different
On this point, WA seems to be exactly as Google, having to build a (guessed) big database and (guessed) big web sites to centralize all the logic to find answers and serve them to people.
Even worse, as explained by Spivack, the tool has to be bootstrapped and fueled by an army of persons entering tons of information by hand only for the sake of WA and maybe a ever growing total paycheck.

3) Monetizing different
On this point, no information is given.
Jon Stokes made his best point on this: even if the process is (sometimes) painful to find an answer with Google, we humans have quite been mastering the process and are able to go through it quite easily.
So, beyond the possible engeeniring masterpiece, what can WA be useful for?
As I explained, the biggest limit of Google is not for humans to find answer but that computers are completely unabled to use it in order to act for humans.

So what may be the most valuable part of WA is NOT THE HUMAN NATURAL LANGUAGE INTERFACE but the COMPUTER ABSTRACTED INTERFACE.
This could be opened and monetized to any other computer in order to get the answers and act upon them.
To take an example back from Spivack, it's pretty easy for an human to get the price of a share from Google with google.com but pretty impossible for a computer provided that a human read through the cumbersome functions written in XML to tell the computer to do so SPECIFICALLY for this question.
As far as I understand WA, it can contain a kind of a way for a computer to ask GENERALLY, making computers as empowered as we are with search/answer engines.