Disqus - Latest Comments for Matthew Gertner

Re: Bob Cringely does his Dorothy impression

Matthew Gertner — Sat, 09 Feb 2008 09:08:29 -0000

Matthew,

My point wasn't to eulogize Cringely. I went out of my way to imply that he may well be talking out his a**. His first paragraph just set me off because I had already been thinking about how many me-too stories there are on Techmeme that bring little to the party.

That said, I can't deny that I enjoy reading his column, even though I can't really argue with your criticisms either. For one thing, he writes well and amusingly. Maybe I'm a soft touch but I thought the whole Wizard of Oz thing was quite clever. Also, he does tend to come up with original and thought-provoking viewpoints, even if half the time I'm jumping up and down on my chair screaming about what a bonehead he is to believe this-or-that nonsense. In fact, that part is half the fun.

Matt

Re: Bob Cringely does his Dorothy impression

Matthew Gertner — Mon, 11 Feb 2008 06:04:04 -0000

No offense taken. Oh, and sorry about misspelling your name. :-)

Re: More on Powerset, the secretive search engine

Matthew Gertner — Thu, 05 Oct 2006 12:17:19 -0000

I'm very much in agreement with Danny on this one. The problem with natural language querying isn't so much that English is idiomatic. Understanding any natural language is a huge challenge because they are ambiguous and imprecise, requiring real-world knowledge to interpret. Take the classic example: "Fruit flies like honey." You need to know something about fruit flies and honey to understand that the ambiguous word "like" is being used as a verb rather than a conjunction.

Despite the enormity of the challenge, over-the-top predictions have been made since the 1950's about the imminent ability of computers to interpret natural language. This is why the term "artificial intelligence" ended up with a bad rep (though this has been tempered somewhat in recent years as computers have finally started to do some stuff which can reasonably be described as intelligent).

So it's certainly correct to be skeptical about Powerset's claims. It's much easier to understand language in a limited domain than in the wild, woolly world of the web, and I can easily imagine them wowing folks with a few canned queries that they know work well. This is very different from performing well in a real-world environment where people are entering all kinds of crud and expecting to get back useful results.

That said, it's always possible that these guys have found a clever way to achieve substantial improvement in search result quality using NLP, without having captured the Holy Grail of actually understanding queries and webpages, boiling them down to some formal semantic representation and matching them up on that basis (which frankly isn't possible today if for no other reason than that our computers aren't nearly powerful enough). If so, that's huge... Google huge. But I, for one, will believe it when I see it. :-)

Re: Cory Doctorow Could Use Mozilla Help

Matthew Gertner — Mon, 10 Sep 2007 08:20:03 -0000

Having gone through the process, I really think that a Firefox bundle with all that implies is total overkill in this case. Cory should do an extension bundle using CLEO (https://addons.mozilla.org/en-US/firefox/addon/...) that includes his theme, FoxyProxy and whatever other extensions he wants to include.

Re: The Coming Search Advertising Crash

Matthew Gertner — Thu, 09 Mar 2006 11:55:40 -0000

Dave totally nails it. The appeal of AdWords (i.e. search engine adds) is hugely greater than that of AdSense (i.e. semi-relevant graffiti). What's more, the problem of clickfraud is tied much more closely to the latter. When you think about it, the idea that anyone can set up a website and make money from people clicking on ads is far too easy to game. Making real cash money is a much bigger motivation than some nebulous idea that you are hurting your competitors by click their links, which is about the worst thing you can do with AdWords.

Wired had a great article about this recently (too lazy to dredge it up) in which they postulated that a botnet could bring the whole AdSense edifice down.

The trend is, IMO, the opposite of what you propose. It's AdSense, not AdWords, whose days are numbered.