obviously referring to a dating site — currently it’s restricted to MIT and Harvard students. This one tries on an idea that I’ve heard discussed numerous times in different contexts, but apparently nobody went and did it in all these years. Instead of running a matching algorithm, it asks third parties (i.e. matchmakers) as well as the interested parties themselves to suggest matches. The thing that is supposed to keep this low-risk is anonymity: a match isn’t revealed until the two primary parties involved mutually accept or their lists intersect.

As with all things that involve anonymity, this asks for trollish and antisocial behavior. I’ve already registered three aliases on moira for exactly this purpose — ok, ok, so they’ve suppressed that antic after people raised concerns, though these and other ramifications should have perhaps been worked through a bit more carefully pre-launch.

The spam potential remains. A matchmaker’s identity isn’t revealed unless both people accept her suggestion, so pranks and insults can be conducted to an extent. One way around this may be grafting social graph data onto the system for collaborative filtering (if they manage to obtain such data…). And if they do, perhaps the suggestions of more closely related people should weigh more, along with those of successful matchmakers. Perhaps there should even be more weight if multiple matchmakers concur. This is extremely intriguing, because eliminating spam is equivalent to predicting who is a likely match, and collaborative filtering for this problem is an unexplored direction.

The more fundamental question is why such a site is even necessary.

Ostensibly, there is a gain over the serial nature of asking in person, due to the ability to make more informative decisions by using data you don’t have or cannot socially obtain in the open. If anonymity compels people to provide more preference information into the system than they would otherwise do, then everybody is better off. This is the positive aspect.

Nevertheless, if we wanted to use full information, why not run a global algorithm? If humans had no feelings, they could just make a list ranking whom they liked in order, and let a computer take care of allocation, almost like housing assignments. But alas, despite the rationality and efficiency of this obvious method, real humans appreciate neither the results nor the implications of it. Nobody likes to ponder the idea of not being #1. So something like nchoosetwo is a compromise, and hides the negative aspect of knowing too much: hurt feelings. But now the site becomes very dangerous. Under the cover of anonymity, the site is collecting ranking information about people from every action on the site. One particular situation in which an explicit rank order is elicited is where there are multiple matches. By your choice, you reveal to the system that “A is better than B.” Is this something the site should know? Not to mention that proposing matches in parallel, when combined with side information from the subsequently unfolding real world, also leaks preference information to an anonymous matchmaker. Those are much more dangerous information than who your Facebook friends are…

So far though, this site has embarrassingly few features. When you have a mutual match, all it does is to print:

Mutual crush! How about a date?

Tellingly, you can’t remove this.

The story goes that on November 30, 2003, he was asked to write some code for an on-campus dating site called the Harvard Connection. Then on December 7, 2003, Zuckerberg wrote in private communication:

Check this site out: www.harvardconnection.com and then go to harvardconnection.com/datehome.php. Someone is already trying to make a dating site. But they made a mistake haha. They asked me to make it for them. So I’m like delaying it so it won’t be ready until after the facebook thing comes out.

Although it isn’t clear to me that Zuckerberg did not conceive of the “facebook thing” long before his meeting, for somebody who claimed he did not want to work under anybody else, it seems at least curious why he would accept spending time to write code for a similar site. On the other hand, he was skeptical about a straight dating site, designed facebook to be not primarily for dating but something more stealthily innocent (the right call). So it is possible that he was mostly concerned with the timing of the launches of two different sites, rather than their content.

In any case, the real shock is that the whole thing took only two months. The facebook domain was registered on January 11, 2004, and the site launched on February 4, 2004.

Sure enough, it was, and you can see it on Youtube.

Some local Seattle event is a bit far from the UK, and sure enough the Daily Mail never reported on this back in the day so far as I can tell. A bit of artistic liberty, surely, but perhaps there is more to it.

Now I’ve always found media subject to money incentives to gravitate toward entertainment. And entertainment being — perhaps — an art form, somewhat like advertising is an art form, the use of images tends not to be about depicting reality, but rather merely to suggest reality, or the reality being conjured up by the writer with words, an analogic reality. Images are nowadadys understood like visualizations on Powerpoint. Fire up an image gallery like Corbis and pick something that fits the words and job well done. If nobody notices, all the better. That is an interesting phenomenon, especially when it comes to news reporting, I must say.

So here you find BusinessWeek discussing Suzhou as one of the world’s worst places to work. And what do you get? This nice photograph to go along with it. And you say, wow that’s some smog all right. Then, you quickly notice that it doesn’t look like Suzhou at all, but rather awfully like Shanghai, which is a completely different city. And you’d be right, because it is Shanghai. It is a shot of Suzhou Creek in Shanghai! A suggestive photograph, yes. Suggestive of the name, even. But unfortunately not Suzhou. As much not, as a random house on a random Washington Street would not be a valid depiction of George Washington’s erstwhile place of abode. Suzhou being quite polluted notwithstanding, we still do not know what Suzhou actually looks like at all. Great. Analogic reality.

As a certain celebrity makes his annual pilgrimage to his funding sponsors and idolators around the world this month and next, will we see more analogic reality like we did last year? Some instances of this art form are after all more egregious.

But I understand… playing ads too many times in error is of no concern to NBC, of course.

So what’s on NBC’s site? Most shows have only the latest episodes on a time delay (to not preempt live broadcast and DVD sales, I suppose), but a few have all past episodes available. One of these I’ve been watching is Friday Night Lights. Actually never saw any ads for this, but all of Season 1 was surprisingly good. Unfortunately, Season 2 was crap and total BS. Sadly, there is only so much good material for writers to crank out.

But guess what, this is an actual industry. Here is an article that shows, to my great amazement, that people have not only taken this concept to heart to solve the real problem (for spammers and hackers) of automated CAPTCHA decoding by low-wage humans, but they’ve even managed to load-balance the whole thing to reduce latency! What … the hell!

We have several big teams and hundreds of active agents solving captchas, all at one time, especially during daytime in India. The backend of this project involves over 45 powerful, expensive servers communicating with the MySpace site to pull the captchas and then queue them up on this site, and then process the results to push back to MySpace all within 20 seconds per captcha.
…
Sometimes 100 or more agents are waiting at the same time.

Now I’m going to be open-minded here… That’s some entrepreneurial spirit. The reason why 10 years ago I thought this would be a joke was because I didn’t think anybody would put up with this kind of work. Not only was I wrong but now that I think about it, I can’t see how it’s worse than factory work or other kinds of manual labor. In fact, isn’t it better?

Let’s generalize this a little in the way I’ve mentioned in the opening and see if it can’t be applied innovatively on a large scale to other projects. This world needs more jobs.

In recent months, some changes were made to the site with interesting implications. These changes are fairly unique to the Chinese language site but there is something to be learned from them.

First, there is currently just one stored version of the contents on the Chinese Wikipedia site. It wasn’t always so. The very first problem faced by implementors in the early days was the script issue. There are two scripts used by users from different areas, Simplified and Traditional. All characters, regardless of script, are assigned code points in Unicode. But it is not as easy to change between scripts as changing fonts, because there is no one-to-one character mapping between the scripts. It is very nearly many-to-one in the Traditional-to-Simplified direction, however. And because the characters have semantic meaning, the one-to-many conversion can also occur, provided a character-cluster context as little as a word (< 5 characters usually). Therefore, Chinese Wikipedia actually began as two stored versions for each article, one for each script. Since this resulted in divergent articles as different people edit different versions, a project was begun to copy edits from one to the other, at first manually, and later on automatically. Thus was born the automated transcription between the two scripts so that eventually, everything was merged and just one version was stored with a user-selectable automatic script switcher. This worked pretty well for a while.

Because of the one-to-many mapping, the implementors used a word table to convert certain words as units. Because of this tactic, they very easily and effortlessly slid into the role of making a dictionary, without even realizing it. What they did was to incorporate entries for lexical differences between mainland China and Taiwan, notably of foreign loan words in the last fifty years. This, as I will show, is a huge mistake.

But first, a story. I remember when I was still not very used to the Traditional script, I asked on an internal Microsoft Chinese employees list whether anybody had some kind of codepage file (basically a dictionary) to automatically convert Traditional characters to Simplified characters. This is a many-to-one task, so a simple codepage would do. The response I got was that, no, Microsoft only had codepages to convert between "locales," which meant Taiwan, Hong Kong, etc., and Unicode. So I asked, well why can't you stack two of these together back-to-back with Unicode in the middle, in order to convert between the "Taiwan" locale, say, and the "mainland China" locale. Then I got a response from some self-righteous Taiwanese Microsoftie that Microsoft only does things "the right way" and to make that conversion "correctly," lexical translation is required. I said, no, I don't care about lexical differences, stop telling me what my question should be, and answer my actual question: can I make a codepage to go from the Traditional to the Simplified script. I never got an answer to that beyond "Microsoft never makes things that are hacked together like that." Two revisions of MS Office later, there came a Word command called "Chinese Translation," which converts between Traditional and Simplified scripts and optionally "translates" some words, too, so at least that problem is solved. But evidently, because of the way things are in the world right now, Simplified and Traditional scripts will always be associated with specific locales.

In any case, as the conversion tables grew in size, the implementors of Chinese Wikipedia asked the users to help maintain the table, but in one of the worst blunders, they allowed the users to edit the tables without specifically agreed upon rules of what these tables can contain and how these tables may be used. As more articles are added to Wikipedia, sometimes the lexical differences crop up in how to name an article. This is fairly similar to what happens on English Wikipedia and isn't something that its rules can't deal with. However, some discussion pages got fairly heated over the proper naming of things not only in the title but as they appear in the rest of the site. And due to the existence of the conversion tables, people began to simply add entries to them so that as what used to be a simple script conversion became a partial dynamic translation.

Why is this all bad? Because it is a loophole allowing privilege escalation. There is a standard process for creating and modifying content, and resolving conflict of content. As soon as the conversion table enters the picture, the process can be bypassed easily by injecting conflicting content into the table, which is not subject to the same process. (In fact it is difficult to even bring up discussion on the table, because it seems so objective — it seems to be just a “transcription table” even when it is no longer.) Now there are two versions of content. Once locale-specific lexical edits to the table became a precedent, other requests came to distinguish between “Taiwan Traditional script” and “Hong Kong Traditional script” even when it isn’t a script issue but a content issue that ought to be dealt with in the discussion pages of specific articles. Instead, by misidentifying the nature of the issue as a script issue, more conversion tables were created and more divergent content not subject to review were added to these tables, so that now there are five versions (!) of Chinese Wikipedia, and none that does not force the reader to read either a regional variant or a mixture of scripts.

What’s the lesson? It is very critical to check the technical process to make sure there are no such privilege transformation devices like conversion tables, unless they are carefully and narrowly defined, or put under the same agreed upon editorial process as articles.

Edit: This post stemmed from a discussion I started on Chinese Wikipedia (user “TTTT”) to advocate for separating transcription and translation. Those changes were implemented about a year later with the effort of many others, and remain to this day. Theoretically, if you replace the “/wiki/” part of any article’s URL on zh.wikipedia.org by “/zh-hans/” (resp. “/zh-hant/”), you should see the pure transcribed Simplified script (resp. Traditional script) version of the stored article, without locale-specific phrasal translations. Unfortunately, the situation has regressed once again with the introduction of article- and topic-specific exception tables where the two concepts are frequently conflated.