« January 2005 | Main | March 2005 »

February 27, 2005

Blink and Interviewing

On vacation I read Blink by Malcolm Gladwell. He visited Microsoft and discussed the book; that link will give you the basics.

It's an interesting book but I think more about "cool stories" than the more profound, applicable insights from The Tipping Point. I think it has a structural flaw: he explains that people can't explain their snap judgements and when asked to do so spout nonsense, but then relays a bunch of people's explanations of how their snap judgements work. Isn't that inconsistent? Even if we allow that certain experts can understand how their snap judgements are made, there is no easy way to take advantage of this; to be able to make snap judgements about something you need to become an expert at it, at which point better snap judgements will become a by-product of your expertise.

He does have some great stories and some of these are about unconscious bias and how even if people know that an experiment is designed to reveal unconscious bias, and which bias it is, they cannot prevent the bias merely by thinking ahead of time "I will not be biased by factor X or Y".

So let's think about interviewing. Microsoft used to hire a lot of the same kind of person -- I don't mean the obvious racial or gender sameness, but just a certain type of geek with a certain type of personality. So the company worked hard to educate interviewers about avoiding these more subtle biases -- about how loud someone spoke, how they held their hands, where their gaze landed, etc.

Now I completely believe what Gladwell has written and reported on in various place: interviews are decided in the first fifteen seconds and the interviewer makes a "blink" judgement which he or she then spends the rest of the time justifying in his/her mind. The brainteaser questions and all that are just hooha. William Poundstone, in his book How Would You Move Mount Fuji? also claims that the brainteaser questions are hooha, but for a different reason, which is that they are too random and "spoilable" if a candidate knows what is going on.

Anyway, I think they are (almost entirely) hooha, but I keep doing interviews, and my rationalization to myself is that I may be incredibly affected by subtle behaviors in candidates, but that may be a good thing; someone who has the proper geek mindset to work at Microsoft is probably throwing off a lot of unconscious signals about their geekiness, which I am picking up unconsciously (I should emphasize I am not talking about obvious traits such as race and gender -- I sincerely hope I am not being affected by those, but if I am, I realize that it is wrong wrong wrong).

One thing I do during interviews is to note to myself what my initial fifteen-second impression of a candidate is. I figure if I am going to be so affected by something, I might as well note it explicitly to myself. Sometimes, by the end of the interview, I find that my initial impression has been contradicted by how the candidate responded in the interview. This can go either way (a candidate I initially liked doing badly, or someone I had doubts about doing well), but in either case I feel like I have done something heroic -- I have overcome my initial impression and undoubtedly gotten the "correct" assessment in the end.

Now this is where Blink made me think. Gladwell says that these initial impressions can actually be more accurate than more detailed analysis, and that having more information about something makes our judgement worse -- while simultaneously increasing our confidence in the validity of our judgement. This would imply that every time (or most times) that an interviewer changes their initial impression of a candidate, they are actually changing it to the wrong impression, while simultaneously increasing their confidence in that impression -- thus making it more likely that this kind of "error" will happen again.

Poundstone's final assessment is that Microsoft-style interviews are not any better than other methods, but they aren't any worse either. But is it possible that all the training Microsoft has done to prevent interviewers from being affected by subtle biases, which encourages interviewers to go against their initial impressions, has actually made the interviewers worse?

Posted by AdamBa at 11:29 PM | Comments (4) | TrackBack

Polynesian Voyaging

I just got back from a week in Hawaii. We stayed at the Grand Wailea Resort, which is everything Caitlin Flanagan said it was.

On the plane over I read an interesting article in the Hawaiian Airlines magazine, about ocean-going sailing canoes. Starting in the mid-1970s with the construction of Hokule'a by the Polynesian Voyaging Society, there has been a revival of the traditional long-distance canoe voyages between Pacific islands. The PVS Voyages page lists some of the journeys undertaken. Tahiti! Easter Island! This is great romantic stuff: "the ancient art of guiding canoes by the celestial bodies and ocean swells". It gets better: the art almost went extinct, based on this quote from another part of the site: "In 1973 there was only one deep-sea navigator that PVS knew of; today there are nine, with several more in training, along with 135 experienced deep-sea sailors in Hawai'i alone-ensuring that the Hawaiian people will never again lose their traditions of voyaging and navigation."

In 1995 two of the canoes were in Seattle, but I was unaware of it. I probably heard about it, but it didn't register as anything unusual, maybe the same as those tall ships people sometimes build and sail around. Here's one account of the visit [second item, marked "June 1995"]; OK, a thousand people showed up, not bad.

Posted by AdamBa at 11:06 PM | Comments (0) | TrackBack

Still Here

I am still blogging. Was on vacation for a week, far away (except briefly) from all computers.

Posted by AdamBa at 12:25 AM | Comments (0) | TrackBack

February 15, 2005

Pie in the Sky

The Seattle Times has a weakness for crazy ideas about improving Seattle. If you have a notion on how to speed traffic, beautify the waterfront, or create a postmodern conurbation stretching from Vancouver to Eugene, chances are you can get on the front page of the Opinion section in the Sunday paper.

Last Sunday they outdid themselves with a plan to redo highway 520 through Montlake (if you don't live in Seattle, you may want to stop here). One sure sign that this was an unusually crack-infested plan was the fact that it was accompanied by an editorial explaining why they print these kinds of articles. But the other plans that the editorial refers to -- a bridge across Puget Sound, ferry service across Lake Washington, NASCAR track in Snohomish County -- are paragons of stone-cold reality compared to this latest scheme.

The basic idea is to replace the highway through Montlake with a suspension bridge. The current highway drops a hundred feet or so as it crosses Portage Bay, then bends to aim towards the lake. No problem, just design a suspension bridge that drops a hundred feet and bends at one of the towers (no, really). Then for added effect, from that same tower that has the bend, a third section of bridge juts out sideways and crosses over to Husky Stadium. And of course that means there is a highway interchange also, with the flyover ramps neatly threaded between the suspender cables. The paper's site doesn't have the picture that was included, but you can't really appreciate it without one, so here it is, by Rob Wilkinson of the Montlake Community Club:

Let's assume he really meant a cable-stayed bridge, not a suspension bridge. I still can't quite see how he would get the bend and the interchange. Of course he has to toss out the name Santiago Calatrava, the current pixie dust for anyone seeking to promote a radical bridge design.

Then he points out that the state's proposal to replace the bridge would cost $2.9 billion, but the Millau Viaduct only cost about $500 million, so his plan should be cheaper. Right!! Because the Millau Viaduct had an interchange and a big elevation change and crossed a lake that was hundreds of feet deep...didn't it? Doesn't everyone know that the only consideration when pricing a bridge is the length? By that logic, the Laerdal Tunnel is 15 miles long and only cost $114 million, so we should be able to build a 2-mile tunnel under Lake Washington for about $15 million.

I think if you boil this down to actual reasonable suggestion, he is saying that 520 should cross Montlake on an elevated highway, and that having a connection across to Husky Stadium would improve traffic.

The first I find doubtful. 520 mostly crosses Montlake in a below grade cut, and I would suspect that most people would prefer putting a lid over that to having a highway overheard.

The second is a reasonable idea. The "ramps to nowhere" on 520 by the Arboretum were originally intended to connect to two other highways: the Anderson Freeway, which would have plowed through the Arboretum, and a tunnel to where Husky Stadium is. Of course it would be nice to have this connection; it's easy to propose such ideas if you ignore things like cost, engineering feasability, and the fact that people would go apeshit if you actually tried to put a bridge there.

What truly annoys me is the attitude of "the Department of Transportation is hidebound and can't really be creative in solving problems, so we need geniuses like us to come up with clever solutions." It's ISTYS syndrome all over again. Funny that the paper would print things like this and then wonder why nobody trusts government and people are losing their faith in the media. Just a coincidence, I'm sure.

Posted by AdamBa at 10:35 PM | Comments (4) | TrackBack

February 12, 2005

The Gift for the Person Who Has Everything

I've been meaning to post this...I bought my wife a gift for the holidays that I think was pretty nifty, if I do say so myself.

It was from a place called Big Fun Comics. I saw a little teensy ad for them in The New Yorker (I knew that rag would come in handy someday). What they do is create a custom comic strip about a person or family (or anything, I guess), based on details you provide. The comic can be black and white or color. I chose black and white for that old-time, Pogo-esque feel (also it was cheaper). I supplied a dozen or so facts about the lovely gal I married, and Adam Holzman (the artist) whipped it into a most excellently risible concoction that was roundly acclaimed as the gift of the millennium (so far).

I highly recommend it for anyone. Adam called me several times to clarify exactly what I wanted and get a couple of things for the kids to say (he carefully worked them into a few of the panels). They also framed it and Fedexed it to me so it got here in time. It works for almost anyone: loved one, boss, child, teacher, etc. They also do specials like custom Valentine's Day cards. Sorry, it's too late to order them for your sweetie this year, but keep it in mind for next.

Posted by AdamBa at 10:00 PM | Comments (0) | TrackBack

Kanji Backspace, Again

Larry Osterman posted a review of Find the Bug. This inspired a Microsoft blogger (who shall remain nameless, unless he/she chooses to reveal him/herself) to buy the book. The first example he/she looked at was the Kanji Backspace one, and naturally this led to an email claiming the example in the book was wrong, and an ensuing discussion among Larry, myself, and Mystery Blogger (not his/her real name). I won't reveal the outcome of the discussion, but once again this programming questions reveals itself as one of the most deceptively hard ones out there -- which makes it all the more amazing that as I related in my first book (and again here), someone once told me that they used Kanji Backspace when doing on-campus interviews of college seniors, and refused to fly someone back to Microsoft for more interviews if they didn't nail it.

Actually I was on an interview loop recently where someone asked a candidate to do strlen() of a DBCS string (the actual Windows stdlib function that does this is called mbslen()). This basically is asking someone if they can go forward in a DBCS string -- which in the "old school Microsoft" context of asking someone about Kanji Backspace is assumed to be so trivial that it isn't even asked.

Posted by AdamBa at 09:47 PM | Comments (1) | TrackBack

35 links from 23 sources

There's something strange about Technorati. When I navel-gaze there, it seems to get stuck on a certain number of links and source. Currently it's 35 and 23 -- it's been that way for weeks, even when some of the links are very recent. It's like they've only got room for that many links and when a new one gets added, an old one falls off. ??

Posted by AdamBa at 09:35 PM | Comments (2) | TrackBack

February 10, 2005

Single-Letter Domain Names

This article about single-letter ticker names made me think about single-letter domain names.

I know these used to be a hot commodity, but a few random checks revealed most are not around anymore. So I wrote a script in Monad to check on them:

[int]$n = [char]"a"
while ($n -le [char]"z") {
    trap [System.Net.WebException] {
    $c = [System.Convert]::ToChar($n)
    $url = "http://www." + $c + ".com"

    $a = new-object System.Net.WebClient
    $s = $null
    $s = $a.OpenRead($url)
    if ($s -is [System.IO.Stream]) {
        write-host ($url + ":")
        $sr = new-object System.IO.StreamReader $s
        while (!$sr.EndOfStream) {
            $l = $sr.ReadLine()
            if ($l -like "*<title>*</title>*") {
                write-host $l

    $n = $n + 1

It gets the contents of the site, if it exists, and then scans for the <title> line and prints it (there may be easier ways of doing this, but it seems to work). The results:

<title>Qwest Communications phone company with Internet, wireless and VoIP service</title>
<title>PayPal - Welcome</title>
         <title>Nissan Vehicles -   2005 Z&reg;</title>

These are the same three that J-Walk found a few years ago. Here's a bit of discussion about the single-letter domains -- they used to be obtainable, but now they can only be renewed, no new registrations.

Posted by AdamBa at 09:48 PM | Comments (7) | TrackBack

February 08, 2005

Word of the Day - Marketecture

Someone used this word in a presentation the other day -- a cross between marketing and architecture. It was in the phrase "marketecture diagram", which is one of those diagrams you'd recognize if you saw it -- lots of colored boxes connected by big fat unlabeled arrows.

In some people this word may induce the same metal-on-chalkboard feeling as neologisms like "guesstimate" and "advertorial". I had never heard it before, but some web searching reveals it is quite common. It appears as a trademarked term, and here is a discussion of the difference between marketecture and tarchitecture. I guess "architecture" has a distinct enough ending that you can slap it on the end of anything and convey the meaning -- blogitecture, webitecture, fooditecture, etc. (I made all those up, but sure enough they all return MSN Search hits).

Posted by AdamBa at 08:54 PM | Comments (0) | TrackBack

February 07, 2005

Microsoft Suckiness

An article title "Why Does Windows Still Suck?" recently appeared on SFGate, and thence on Slashdot.

The title is a bit misleading; the events happened a year ago (before Windows XP SP2 shipped), and even then the author describes his SO's computer as a "creaky Sony Vaio PC laptop". So she was probably running -- what -- Windows 2000 or something? Well, no surprise there. The same thing happened to me when I put a Windows 2000 machine out on the wilds of the Internet. I had to do a clean install of XP SP2, but now the machine seems fine on the Internet. Buy Viagra now. Well, almost fine. No, really, I have not detected any spyware on the new system.

The article purports to be fair and balanced, but of course it isn't. He uses an analogy with a car (common target of bad computer analogies): "Here is your brand new car, sir. Drive it off the lot. Yay yay new car. Suddenly, new car shuts off. New car barely starts again and then only goes about 6 miles per hour and it belches smoke and every warning light on the dashboard is blinking on and off and the tires are screaming and the heater is blasting your feet and something smells like burned hair. You hobble back to the dealer, who only says, gosh, sorry, we thought you knew -- that's they way they all run. Enjoy!"

Of course it's not like that. As he said, the computer worked fine before she put it on the Internet. It's more like, you get a brand new car. Works great. Then you take it to a war zone and drive down the middle of the street. The tires get punctured, the windshield cracks, the engine explodes. Yes, the car stops working.

Sure, Microsoft continued to sell convertibles when it should have been selling up-armored Humvees, but the fact is it really is a war zone out there. But the article glosses over the real question, which do Macintoshes have thicker steel, or do they (for whatever reason) have virtual red crosses painted on them? The article states: "And I know, finally, the argument that says that if the world was using Macs instead of PCs, the hackers would be attacking the Macs...Which is, of course, mostly bull. I'm no programmer, but I know what I read, and I know my experience: the Mac OS architecture is much more robust, much more solid, much more difficult to hack into. Apple's software is, by default, more sound and reliable, given its more stable core." Since I worked on the Windows core I am pretty confident that it is just as reliable and stable as Mac OS X; the vulnerabilities you see exploited are mostly not in the Windows core, and if they are, they are not due to any fundamental issue with the architecture of Windows, but are just plain bugs.

As the author says, he has no way of knowing this; his end-user experience can't really give him direct insight into the architecture of the OS, only empirical evidence. At its heart, his argument is circular; the Mac has fewer attacks because it is more robust, and the proof of its robustness is that it has fewer attacks. It reminds me of the special shoes I wear, the ones that keep crocodiles away.

Posted by AdamBa at 10:17 PM | Comments (8) | TrackBack

February 06, 2005

Big 'Bucks

Yesterday I went to Starbucks and bought a mocha for me and a hot chocolate for my son. Look at the receipt on the left. It cost $5.90, before tax, for that.

$2.25 for the hot chocolate is particular amazing. The raw ingredients for that are 8 oz. of milk and a squirt of Hershey's chocolate syrup, plus a cup and a lid. That can't possibly cost Starbucks more than 10% of what they are charging me for it.

They have rent and employees to pay, but I have read that a typical restaurant has to charge 3 times what the ingredients cost to cover all the overhead. Here Starbucks is charging more like 10 times what it cost.

Microsoft also charges a lot more for software than the actual cost of manufacturing. It's not clear what it actually costs, but it can't be much; I'm pretty sure that duplicating a software CD costs the same as duplicating a music CD that sells for $10. Microsoft employees can buy discounted Microsoft software at the company store; the cost was always described as "COGS" (cost of good sold). Office Professional costs a little more than Office Standard, which made some sense back when Office Professional came on more floppies than Office Standard. Today, they both come on one CD, but Professional is still more (the current company store prices are $50 for Office Standard sells and $60 for Office Professional).

The difference is that Microsoft has huge up-front development costs for Office; I don't think there is much development costs in figuring out how to mix chocolate syrup and warm milk to make hot chocolate.

Posted by AdamBa at 09:38 PM | Comments (8) | TrackBack

February 03, 2005

MSN Search Launches

As you may have noticed, MSN Search has officially launched.

The search results seem to be good (the answer on launch day was #12 and #1; I won't repeat the question). I'm not crazy about the UI, however. I guess it is part of MSN branding but the blue box around the search box is oppressive and I dislike the white on dark text. I also don't like the way the results page looks. Basically the fact that the page titles are boldfaced (so the search term can't be highlighted by boldfacing), and the blue bar at the top again. So my summary is that I like all the places where it is just like Google, and I am perturbed by the places where it is different. Edward Tufte commented that Google's UI was perfect, and he may be right. Or perhaps I am so used to Google that I view anything else as "wrong". But the fact is it makes the search results seem more amateurish, even though (at a quick navel-gazing glance) they are actually better than Google's.

Posted by AdamBa at 10:20 PM | Comments (1) | TrackBack

David Weise

Larry Osterman blogged about David Weise leaving Microsoft.

When I was a senior in college, I very much wanted to work for Microsoft. I carefully noted the date in January 1988 that Microsoft was coming to Princeton to do on-campus interviews, and on the first day you could sign up for interview slots, I rushed down to Career Services to get one (companies coming to Princeton usually either allowed first-come-first-served open signups, or else prescreened resumes to choose who they would interview. Microsoft used open signups).

A fellow student who had interned at Microsoft the previous summer told me that Microsoft was having a dinner for former interns (who all had guaranteed full-time job offers) and they had said he could invite anyone else he felt would be interested. Of course I was interested, so I cadged an invite. Microsoft was taking us to Lahiere's, the fanciest restaurant in town, and better yet was footing the bill.

Two of the Microsoft employees present had PhDs from Princeton: Nathan Myhrvold and David Weise. This was one of the first employer-paid meals I had eaten, and I debated whether ordering caviar would compromise my chances of being hired. Nathan Myhrvold, bless his gourmet heart, solved the problem by ordering some for the table.

The actual interviews were the next day and David Weise was my interviewee. These were the quick half-hour interviews that decided if Microsoft would fly you back to Redmond. Microsoft was famous back then as the only company that asked candidates to write actual programs during these interviews, and sure enough David Weise asked me to write the code to computer the Nth Fibonacci number. I managed that with no trouble, but then he asked me what the running time was (as in, O(N log N) etc). I flailed at this for a bit before he told me that the answer was O(Fibonacci). Yeesh. Despite this, I did get the flyback to Redmond (but didn't get the job that time).

The other thing I remember about the interview on campus was that I was wearing a tie and the first thing David Weise said was that he was glad he didn't work at a company that made him wear a tie because it cut off the blood flow to your brain.

Posted by AdamBa at 09:32 PM | Comments (1) | TrackBack