Wednesday, August 1, 2007

Things I might be interested in

In February 2006, a friend of mine moved from California to New York city. To make the transition easier, I gave him the Zagat New York City restaurant guide, which I purchased through Amazon. Since that day, "As someone who has purchased or rated books by Zagat Survey", I have been notified about every two months of new issues of Zagat guides, covering the most random places in the US. I have been suggested to purchase the Zagat 2007 Texas. I have been advised to look into the Zagat survey of America's top golf courses. And today I have been recommended to check the 2008 Zagat survey of Washington Dc/Baltimore Restaurants.
Granted, it's notoriously difficult for computer to make recommendations based on taste, but come on! This is just plain ridiculous. I haven't heard anything from Amazon on the latest issue of the San Francisco Zagat (to be fair, I may have forgotten), even though they have my address on file - so why would the computer decide that I may be interested in Baltimore, Texas and US golf courses, when none of my amazon purchases has anything to do with these?

Labels: ,

Tuesday, July 24, 2007

Making humans more computer

Sorry if this article is last month's news, but I still find it interesting. There is something pleasantly twisted in trying to make computers more human, and ending up making humans more like computers... And it is somewhat reassuring to know that people are still so much better than computers - but that if you want to see any work done, you'd better make it fun. At least, when the machines take over in the future, they'll have to keep us entertained.
clipped from
But reCaptcha has an even sneakier — and more delightful — purpose. The words are pulled from the book-scanning project of the Internet Archive
One of the two words in the test is the control word: The gatekeeper computer knows what it should be, so it's there to make sure the puzzle-solver is indeed human. But the other word is there for a different reason. The Archive's scanners are good, but some of the words are too smudgy for the software to decipher. The game takes the image of each smudgy word and puts it into reCaptcha. Each time someone completes a reCaptcha puzzle, they'll be doing a tiny bit of work — translating that difficult image into text

Roughly 50 million Captchas are solved each day. If von Ahn can acquire just a fifth of those users, he'll have a stunning 30,000 daily man-hours of work at his disposal. It would constitute the world's fastest and most accurate character-recognition computer, processing 10 million words a day.

 blog it