Slashdot Mirror
Wolfram Alpha vs. Google — Results Vary
wjousts writes "Technology Review has an article comparing various search results from Wolfram Alpha and Google. Results vary. For example, searching 'Microsoft Apple' in Alpha returns data comparing both companies stock prices, whereas Google top results are news stories mentioning both companies. However, when searching for '10 pounds kilograms,' Alpha rather unhelpfully assumes you want to multiply 10 pounds by 1 kilogram, whereas Google directs you to sites for metric conversions. Change the query to '10 pounds in kilograms' and both give you the result you'd expect (i.e. 4.536 kg)."
Karma be damned, but..
No one cares about a new search engine. Really, Google suits all my needs.
Funny, but it could be improved. I tried to get Google to convert decimal to hex the other day, no luck. It also doesn't convert radioactive decays per minute (dpm) to microcuries. These would be useful. Not that I can't do it myself.
Give me Classic Slashdot or give me death!
Alpha is meant to interpret natural language to figure out an answer. "Microsoft Apple" and "10 pounds kilograms" aren't natural language questions or common phrases. Those would be keyword searches, which is what you'd type into Google. Try "Compare Microsoft to Apple" or "How many kilograms are in 10 pounds" and you'd be using Alpha more appropriately.
Each system is a tool. If you don't use the tool as described you won't get the results you're looking for.
Developers: We can use your help.
I've never heard of Wolfram Alpha, so I googled it. Then I thought: If this new search engine becomes popular, will I still use google as a verb? I'd hate to wolfram stuff.
Stupid "face off" story.
WA doesn't compete with Google.
WA works with structured data sets and natural language queries to come up with replies, Google searches the web. WA won't do shit with a query like "digital camera reviews", but Google will. Google won't do shit if asked to calculate answers based on statistics, WA will.
Beware: In C++, your friends can see your privates!
Does Wolfram do any better than google when you type "hot free porn videos". Will you be able to type "teenage pussy" without being bothered by some old deary who wants to tell you about the longevity of her pet cat?
Typing in "Cancer New York" could mean anything.
If you gave that question to a human they'd have no idea what your were looking for.
Why didn't he try asking the question he was trying to ask which was "What are the rates of cancer in new york?" or even just "Cancer rate in new york"
All his other searches are equally stupid.
"Some days even my lucky rocketship underpants don't help."
Reminds me of when I was in France, and still having trouble understanding the spoken French language. I was talking to a guy who asked me, in translation, "Brothers, sisters, one, two, three?" It took me a while to figure out he wanted to know how many siblings I had. Dumbing down the question like that didn't help me understand him any better, it made it worse. Using correct French grammar and simply slowing it down would have been much more helpful.
I imagine Wolfram Alpha is like that.
Is Wolfram Alpha especially good in doing side to side comparisons (ex. from the article: "Microsoft Apple", "Stanford Harvard", "Utah Florida", "Utah Florida population")? Or why did the article test both engines with those queries?
I would have rather expected, complete questions that are nevertheless hard to answer (unless you know a source), such as:
1) "How many bull terriers are in the UK?"
Google: link to Bullterriers on Wikipedia and some dog clubs in the UK.
Wolfram: ???
2) "How many blind people live in the US?"
Google: first link to WikiAnsers (about 1 million, but without any references). Next links seem to be more serious, but difficult to get a real answer to that question (it depends on how you interprete "blind").
Wolfram: ???
3) "What is the color of a strawberry?"
Google: This confuses me, apparently it has many colors...
Wolfram: ???
4) Apparently we need to use a comparison question: "strawberry blackberry" ...
Google: I am getting hungry when I am following all those recipe links
(due up tomorrow)
Some might say that Mathematica, the source of my fortune, and A New Kind Of Science: A Brief History Of My Stupendous Intellect were ambitious projects. But in recent years I've been hard at work on a still more ambitious project: Wolfram Alpha.
Fifty years ago, people assumed that computers would quickly be able to handle all kinds of question. It didnâ(TM)t work out that way. But a few years ago, I realized that I was finally in a position to do it myself. As I'd always expected I'd have to, of course.
I had the crucial ingredients: Mathematica and A New Kind Of Science. And my truly massive intellect. With these, I had a language to compute anything and a paradigm for complexity from simple rules. And my spectacular brain, which is much more spectacular than anyone else's, as proven by me being rich as well as smart. Which is smarter: to be a professor, or to have all the professors pay you tribute? I think my net worth makes the answer clear.
But what about all the actual knowledge that we as humans have accumulated? I realized we needed to make all data computable as knowledge. Of course, natural language is incredibly difficult for computers. So we added the secret ingredient: my jaw-droppingly spectacular brain, undoubtedly the largest on Earth.
I'm happy to say that with a mixture of clever algorithms and heuristics, linguistic discovery and curation, and some casual Nobel-worthy theoretical breakthroughs in my spare moments, we've made it work. Itâ(TM)s going to be a website with one simple input field that gives direct access to my superlative brain, in its planet-sized glory.
Our pre-launch testers have been at work as well, and I'm dealing with all manner of queries in spare thought cycles while I jetset around the world, wowing the pitiful minds of gorgeous international supermodels before impregnating them with my superior genetic material. Let's just have a look at the query stream: "tits" "goatse" "mary whitehouse naked" "4chan" "tubgirl" "2girls1cup" "ITS OVER 9000 LOL" "desu desu desu desu"
ERROR ERROR ERROR ####(^^(856*##&##
NO CARRIER
http://rocknerd.co.uk
Wow, is it really that wrong?
How about a test involving actual English-language questions and not just keywords? You know, like all those old tests from school that said "please use complete sentences". There is a reason languages have things like prepositions, adjectives and other parts of speech. They actually help put your communication into context.
Nobody knows what the hell you mean with "Cancer New York" because there is no context. How about "cancer statistics for new york" or "cancer treatment in new york"?
Learning HOW to think is more important than learning WHAT to think.
For the Americans in the audience, 1 £ kg = 3.33 $ lb.
I recommend not saying this aloud for it sounds very silly.
Alpha rather unhelpfully assumes you want to multiply 10 pounds by 1 kilogram
Actually, while I agree that is unhelpful, I also don't like the other assumption. Maybe I'm already growing old, but I don't mind if people actually say what they mean instead of speaking or writing in some kind of shortcut-verbs-are-too-expensive-so-I-leave-them-out abbreviated style and leave it to the listener/reader to decypher whatever it could possibly be they mean.
So if you want 10 pounds in kilograms, what exactly is the trouble with actually writing those three (counting the space) additional characters?
Assorted stuff I do sometimes: Lemuria.org
I expect something from Wolfram like the answer Google gives to this question:
How old is Demi Moore?
http://www.google.com/search?hl=en&q=how+old+is+demi+moore&btnG=Search
At the top, you'll see text that says:
Demi Moore -- Age: 46 years (born November 11, 1962)
According to: (some source) [more sources]
This is the proper way of answering a question like that. I don't want just the answer. I want to know where the answer came from.
How many french died at the Battle of Agincourt?
I expect a number from Wolfram Alpha, as well as a cited source. There could also be, like Google, the option to choose other sources.
Eventually, this will all boil down to me driving in my car and saying, "Computer. Tell me: At what speed did Marty McFly need to drive to travel in time?"
Type this into google:
Who is Jamie Lee Curtis' mother?
Look closely at the first entry.
---Technology will liberate us if it doesn't enslave us first.
Most search systems do well if you can find the "magic word", the word or phrase which nearly uniquely describes what you're looking for. When you're searching for something which is described with common words, not terms of art, search engines usually don't do well. That's where to test Alpha, which supposedly has some "understanding" of its data.
Here's an example of something I was doing today. I'm sketching out a design for a special-purpose DC-DC converter, something I haven't done before. I'm looking at a data sheet for a transformer, and at the rules for describing a transformer to LTSpice, a circuit simulator. LTspice wants a value K, the "coupling coefficient". The data sheet for the transformer has various numbers about the transformer, but not the coupling coefficient. How do I calculate the coupling coefficient?
It turns out that the magic words for answering this question are "leakage inductance". Once you know that, you can find the Wikipedia entry that gives the necessary conversion formulas, and calculate the coupling coefficient. Until you find the magic words, though, it's tough. If you just go looking for "coupling coefficient" in Google, you're directed to theory papers. "Leakage inductance" is the number that appears in data sheets, because it's directly measurable.
If Alpha can answer questions like "How do I compute the coupling coefficient for a transformer given the data sheet parameters?", it will be a nice capability.
Yet, if you put in the entire phrase "How do I compute the coupling coefficient for a transformer given the data sheet parameters?" as a query to Google, you get as a first result a paper on how to model a transformer in LTSpice given data sheet information, which is exactly the right result to return. The answer is in that paper, and it's a good paper. Google does better at this than one might expect.