Tell HN: Wolfram Alpha preview goes live

Some caveats, by the way:

- 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.

How often do you need an answer to something that is not already available on the web and there is no other tool on the web that can compute it?

While 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.

And what do we, as a startup crowd, keep saying about niche markets?

I don't know. What do we say about niche markets?

Niche markets pay my bills, so I'd say they're great.

And of course, you've come to this opinion after extensive use of Wolfram Alpha over a good period of time. No? Just empty prejudices based on second-hand impressions?

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.

In addition to second-hand impressions, there's a long history of failed products/approaches similar to this one.

Google has found a sweet spot. And it won't change anytime soon.

which are the the similar failed stories?

Natural language interfaces have failed repeatedly. And so have attempts at the semantic web.

What does work and scales is more free form as is the case with the Web today.

And Google makes "free form" work.

there are people out there that cant write in C. Does that mean good programs wouldnt have been written in it?

Remember Cuil? Fostering press that you're an X killer is a good sign of future failure.

you havent even tried it yet. How is it possible το compare two different products? You can put two different people implement the same concepts in algorithms and you will get two different results and you compare a whole product because others failed?

I can appreciate Empiricism as much as the next guy, but Google Search, Cuil, and Alpha seem comparable.

edit: the dreaded homonym! I mean they can be meaningfully compared to each other, not that they are the equivalent.

Read the thread again, I was simply giving an example of a similar failed story. Wolfram may or may not fail at this, I simply believe hype of a Google killer before launch is a bad sign.

Your criticisms apply just as easily to the carefully built, optimistic buzz Wolfram is attempting to create...the same thing he tried with "A New Kind of Science".

the same thing he tried with "A New Kind of Science".

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.

I hope that was sarcasm

So, what tipped you off?

I think Wolfram Alpha targets a different target audience as Google does. I second the fact that WA appears to be a niche product, since a lot of people will find more information using Google rather than WA (the "Orson Welles" example is a good one. the second search result on Google is Welles' IMDB-page). summa sumarum we have to wait 'til the 18th of May to see what is hiding behind the big WA wall.

All I can say is 'it's different'. Wolfram said this morning that he is not seeking to compete with Google, about whom he spoke generously. Rather, his goal is to provide a fast interface to a wide range of computable data.

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 :-)

We've all been using Google (and web search in general) for so long that it has fundamentally shaped the questions we ask of the Internet to the point that we don't even think questions that we could ask, but don't. We filter them out of our thought process. Just as the people of 1980 didn't keep saying to themselves "Curses, if only it were 2005 and I had the service that will eventually be called Google!" every time they thought of a question. (With a few exceptions, but only a very few.)

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.

Those who are trying to reinvent data search are underestimating Google and the Web at its present form. The fact that a search on any famous film director or actor returns IMDB, or the fact that "When was JFK shot?" (mentioned in another comment here) returns a highly relevant Wikipedia article, tells us that the Web has already adjusted itself to be the semantic Web, and most importantly, a simple and predictable one. In simple cases like I mentioned above I know what to expect from Google, because I know how it works and also the Web constantly fixes itself to be searched easily. So this is a simple, very democratic and predictable system we all use successfully.

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?

You can't do computation with those results. At the first level, this means you have to click a link and load a page. But it also means that you can't ask compound questions ("What films did Orson Welles make that star Ruth Warrick?"). And you can't use the answers in other programs thru an API.

If the question is a commonly asked one, someone will build a service to answer it. And that person could also provide an API.

It seems like you're asking for something like what Wolfram has built.

Actually, what I'm describing already exists. It's the way the Web works today.

Just tried your query and got perfect results in Google, right at the top. And I'm yet to see how WA answers that question. I think it may be able to answer it, but only if it was designed to understand this particular kind of questions and if particular pieces of information were selected for inclusion in WA's database.

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.

That query on Freebase, ready to compose and consume as JSON: http://tinyurl.com/on79u3

The problem is one most people don't know they have, unfortunately. Google only lets you talk about crude keywords to find prebuilt data. If WA lives up to the hype, you should be able to formulate a hypothesis involving arbitrary supported datasets and have WA build you data tables and visualizations.

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.

http://www.google.com/search?q=Fertility+vs.+per+capita+GDP+...

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.

The results for "Fertility vs. per capita GDP for continents" is significantly less useful though.

http://www.google.com/search?hl=en&safe=off&q=Fertil...

True, because neither is the question itself very useful, and thus nobody cared to build an answer for it.

I got some...interesting results by exploring these terms; for example, i was able to get a scatterplot of infection rates vs size of population in Africa. On the other hand, I can't seem to pipe data into a map to save my life. Maybe that's not switched on yet, some things explicitly aren't. More info tonight or tomorrow, after I've tried running Mathematica at the same time.

Sounds like you're thinking of the visualized datasets in http://www.gapminder.org/ and the TED talk http://www.ted.com/index.php/talks/hans_rosling_shows_the_be...

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).

underestimating Google? I place Google pretty high up on the scale of having smart people. Really high. Like #2. Actually, the only people that can beat extremely bright engineers at this game are mathematicians. That's what Wolfram does (remember, Mathematica?).

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.

I was wondering today what the average/median string length is in programming. Google hadn't a clue. I'm curious if Wolfram does have an answer.

You mean, how long is a piece of string?

No. I'm currently implementing a VM, and was wondering whether there's any point in inlining strings <= 3 chars in machine words. Knowing what percentage of strings is <= 3 chars would tell me whether that's a worthwhile optimization.

Oh OK. I thought it was a rather abstracted joke =)

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?

I'm looking forward to seeing if it can answer the question, "When was JFK shot?". Was chatting with someone previously from PowerSet who maintains it won't be able to, but I'm hoping he's wrong...

He's right. 'JFK' gives me: full name | John Fitzgerald Kennedy date of birth | May 29, 1917 (91 years ago) place of birth | Brookline,Massachusetts,United States date of death | November 22, 1963 (age: 46 years) (45 years ago) place of death | Dallas,Texas,United States

Oh, and (like many topics) it also offers a Wiki link in a sidebar.

Actually..

Ask it "Who killed JFK" too.

That question is interesting from a techie AI perspective, but irrelevant for end users IMHO. If they want to know when JFK was shot, they'd probably google "jfk shot" or go to his wikipedia page and search in the page for 'shot'.

Getting users to 'unlearn' that behavior and instead use natural language, and then trust the answer is going to be hard IMHO.

What does "Orson Welles' filmography" return?

Wolfram|Alpha isn't sure what to do with your input.

Like I say, I'm still finding out how to parse queries effectively, particularly for comparison data...in between working (cough).

How about "What is the complete list of films made by Orson Welles"?

How about "What are Google's search results for 'Orson Welles filmography'?"

I guess Alpha could come with tech support to gently rearrange your questions, but it seems like cheating ;)

C'mon, you think I didn't try all those :-)

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.