Tell HN: Wolfram Alpha preview goes live
First impression: useful, but there's a learning curve. If you want a specific datum, it's usually there and presented in a a variety of useful contexts. Sometimes information is obviously there but it's not clear how to access it; for example, searching on 'Citizen Kane' will tell me it was directed by Orson Welles (along with other basic data), but searching on Orson Welles just gives the bare biographical data, not his filmography. It's not obvious how to dig that sort of information out.
The computational aspects are impressive and quite accessible thanks to good NLP; if you've already used Mathematica you'll be familiar with a lot of them. I feel like there's a great deal of power that I haven't properly learned how to manipulate yet. I'm looking forward to the aPI, which (per this morning's Webinar) Wolfram says will be public and IIRC, free.
Paid services will include license of the Alpha software for use on a company's internal datasets, as well as for commercial use of the data and 'large' computational tasks (whatever that means).
Finally, it's open to the public on May 18th.
50 comments
[ 7.4 ms ] story [ 127 ms ] threadWhile Wolfram Alpha is more sophisticated than Google in some ways, Google is better in terms of being practical for a much larger number of people.
In other words, this appears to be a niche product.
Honestly, what's with the battle lines being drawn? The product hasn't come out yet, and basically no one has used it. It seems like it will be an excellent compliment to existing search engines and data sources, but you have to wait and see.
Google has found a sweet spot. And it won't change anytime soon.
What does work and scales is more free form as is the case with the Web today.
And Google makes "free form" work.
edit: the dreaded homonym! I mean they can be meaningfully compared to each other, not that they are the equivalent.
Which is, it turns out, the single most important book about science in the history of the printed word. In fact, it might be more important than the printed word itself. It's a new kind of printed word.
As an example of its quirkiness, 2 slices of swiss cheese + 1 leaf of lettuce + 1 quarter pound burger gives me a USDA style nutrition label. If I add bacon it says it doesn't know what I mean :-)
The question is, will Alpha expand our capabilities and thus expand the questions we would ask? Of course you don't have any questions that could be answered by Alpha but not Google; you've been trained out of asking them for over a decade now. Can Alpha change that?
Beat me. I certainly trend skeptical. But an important part of being a skeptic is being able to change your mind with more data, and I am ready to try it out and see. We can't really have an informed opinion on this question until we live with it for a while, which absolutely nobody has.
Oh, and (like many topics) it also offers a Wiki link in a sidebar.
Ask it "Who killed JFK" too.
Getting users to 'unlearn' that behavior and instead use natural language, and then trust the answer is going to be hard IMHO.
Like I say, I'm still finding out how to parse queries effectively, particularly for comparison data...in between working (cough).
Until I get home (and can fire up Mathematica) I'm not sure how to do such symbolic queries. It's not Deep Thought - I feel like getting the most benefit from it would require some kind of Regex. Might be me. They haven't put the FAQs or tutorials up yet.
I do feel a lean towards computable rather than retrievable data, if that makes sense. I just started with that example as something I knew about that seemed to typify the popular imagination of how it might be used.
After thinking about your query, the only good idea I had was to take something like a large open source codebase and write a program to search for string/char declarations. It'd probably be faster just to try your idea and see whether it helps or not :-/
Oh, and nothing about that on Alpha, sorry. I did ask it 'how long is a string' for fun, and it suggested that the length of a given random string is 12. And who am I to argue?
- it IS a preview: not everything is switched on - it's NOT pretending to offer natural language processing for general queries - they're NOT pretending to be a better Google or Wikipedia - I don't know what I'm doing: I'm in Fisher-Price mode, and there is a learning curve.
I am quite impressed, and also confused.
On the other hand, it is always hard to deal with systems that we have little knowledge about. I suspect WA is the "hard" case.
So, what problem Wolfram is trying to solve? Anyone has any problems with Google?
WA, to my understanding is not entirely self-regulating like the Google/Web bundle and thus it will be difficult for WA to catch up. We'll see.
Will most people realize the sorts of questions they can ask? Probably not. Do most SQL datasets get used to their fullest potential? Probably not. Imagine what it does for the curious student, journalist, or professional, though.
Some things I'd like to see:
1. Percentage of population in African countries with HIV/AIDS over time. Regional, or demographically coded information would be interesting as well. I'd also like it sourced. I could go look this up somewhere... or I could just ask WA.
2. Fertility vs. per capita GDP for countries/regions in the world.
This is not to mention all the historical event, chemical, physical, etc. data.
Imagine NYT data visualizations built on WA data. Private data sets that people slice and dice with WA and then sell specific visualizations, etc.
It's a problem we don't know we have. Success will critically depend on flexibility, ease of use, and accurate and broad data sets.
You can click on the first link - your answers are there.
Notice that the two examples you gave belong to a fairly narrow field - statistics, something that seems to be the trickiest to collect in more or less structured form. But try it yourself, you'll see how easy it is to find virtually anything nowadays. In most cases there's a good chance someone in the world thought about something you are interested in and has put it out on the Net already. If you are the first to ask a question, then I doubt WA will give you useful results either.
http://www.google.com/search?hl=en&safe=off&q=Fertil...
WA seems like a generalization of that. Instead of seeing it as a google-killing AI inference engine, you're saying to think of it as a way to do effortlessly what you could do already... if you went to the trouble to gather the data, put it into mathematica or similar analysis tool.
This is a very credible claim, and I can imagine it quickly finding unforeseen uses, as you say - provided it really is reasonably effortless (as in your four criteria).
The complexity that mathematicians can compute is at least an order higher than engineers. Think of Linus vs. grandma struggling with 3 lines of HTML. The same gap is between the brightest of engineers and a good mathematician.
So that's why I place Wolfram #1 and Google #2, for that particular space (computation). But if you are a bright engineer and never got exposed to good mathematicians, you may just not even be aware that there is all this realm of knowledge above your head.