Blog Post

Cuil Failed at Search, Now Fails to Copy Wikipedia

Stay on Top of Enterprise Technology Trends

Get updates impacting your industry from our GigaOm Research Community
Join the Community!

During the rise of the Beat movement in the 1950s and ’60s, avant-garde writer William S. Burroughs developed a process he called the “cut up” technique, in which he would literally cut out sentences and passages from poems, stories and books (both his own and those of other writers) and stitch them together. If Burroughs had ever decided to automate this process and develop an online encyclopedia, it would probably look a lot like Cpedia. The new offering from Cuil — a startup (pronounced “cool”) that launched in 2008, claiming to have developed a better and faster search engine than Google’s — is destined to do at least one thing very well: make even the most poorly-researched Wikipedia page look like the repository of all the world’s knowledge.

Cpedia launched last week with a blog post from Cuil co-founder and former IBM staffer Tom Costello, who described a meeting he had with Sun Microsystems co-founder Bill Joy when Costello and his wife Anna Patterson (a former Googler) were trying to raise money for Cuil. Joy told Costello that people didn’t need a new search engine that just returned a list of results, they needed something that would write an article based on a search. A note on Cpedia topic pages reads: “We find everything on the Web about your topic, remove all the duplication and put the information on one page.”

It’s important to note that this doesn’t say the service finds everything on the Web and makes sense of it and then puts it all on one page. If what you want are snippets of articles from somewhere (links to source pages are difficult to find) mixed up seemingly at random and then displayed as though they were a coherent encyclopedia entry, even when they are not, then you are going to love Cpedia.

To take just one example, in the entry on Philo Farnsworth, the man who many credit with inventing the modern television, the article starts with a reference to — and a large picture of — an actor named Jimmy Simpson, who apparently played Farnsworth in a movie. There is some history about the development of television and the race with RCA (which reverse engineered Farnsworth’s patents and took credit for the discovery), but it’s all mixed up with references to Simpson and the movie, along with random people including actor Sid Caesar and Jonas Salk, as well as snippets of Farnsworth-related information that appear without any reference to anything.

In his blog post launching the service, Costello says that Cpedia “is very different from a traditional search engine, and not at all like Wikipedia, but that is its strength; it is something new and different.” The Cuil founder is almost certainly right. Unfortunately, being new and different doesn’t necessarily mean that it is either good or useful. Other users who have tried it out describe it as “sentence after sentence of automated nonsense,” and Tumblr and Instapaper developer Marco Arment says that “if this feature is meant to become a serious product, I truly feel bad for them.”

If nothing else, Cpedia proves that there are some things that algorithms and automated processes can’t do — and one of those things is to make sense of all the information that exists on the Internet. Perhaps human beings are good for something after all.

courtesy of Flickr user Kellan

28 Responses to “Cuil Failed at Search, Now Fails to Copy Wikipedia”

  1. Cpedia is a trainwreck. I tried to look up various basic articles, and walked away knowing less before. Plus, the writing style is so monotonic and… unpleasant.

  2. Agree with Greg. All new ideas have to start somewhere and don’t necessarily work as well as one wants them to, but this idea may inspire the next idea which inspires the next and so on. I’m all for seeing people try new stuff even if it fails the first time out. Even Cuil itself was a little novel – its a shame it didn’t deliver, but it was a small step for… er, somewhere :-)

  3. this is the biggest waste of my time…ever. nothing I searched for had any type of result that was even remotely useful. I could cut and paste the google search result w/preview into a word doc and it would crush this lousy pos. I am going to make better use of my time and go scrape gum off the bottom of tables at McDonalds

  4. If you are trying to say (as I think you are) that cpedia sucks, you are probably right at the moment.

    It seems, however, a bit harsh to slam an alpha service so hard, without stopping to give it the benefit of the doubt.

    I think the idea of creating a digest or article based on a search is fundamentally an interesting one. No, the results are not very good right now. But they will improve as the technology matures.

    Yes, humans currently have the edge in doing this. But human labor is very expensive. CPU power is very cheap.

    Yes, the service could and should be better. But wouldn’t it be better to offer some constructive criticism instead of taking a free shot at the groin?


    • Thanks for the comment, Jussi. You are right that this is a difficult problem to solve — and I wasn’t suggesting that Cuil should stop trying to solve it, I was simply pointing out that their current iteration is not good. I will happily report on it when it improves.

  5. You guys…how on earth could it do pages like Wikipedia – this is automatic And its got Alpha on it. I think this could be a really interesting development. Obviously it needs tightening up but what is it with the refusal to allow experimentation? There’s some interesting science going on here.

  6. I checked out Cuil yesterday just to see what happened to it or if it improved any. After my search I could only ask “what the f*ck is this?”. I couldn’t figure it out and I’m sure most others can’t either.

    I never thought it could be possible, but Cuil is worse than it was at launch.

  7. Chancey Mathews

    It’s like an encyclopedia written by someone with ADD. Or someone who just copy-pasted excerpts from search result pages. It’s kind of fun to read. I love it! I can’t wait to see what ridiculous new thing Cuil will give us to laugh at in 2012.

    • Chancey Mathews

      Woha, turns out I was right. On the right of each article page is a link to switch to search results instead: they’re just putting the search result excerpts into an article format. To be fair, their excerpts are somewhat more useful than Google’s, but their results in general are not.

    • Exactly my first thought. Sad news is that Google will index and rank this kind of “content” happily. Specially if it’s fast :-).

      What they(Cuil) don’t seem to get is, it’s much easier to analyze the concept/text flow of an article with a machine then to create one. So hopefully we will get a search engine which will just ignore this kind of “content”. I have no idea why people seem to think that text is keywords.

      Tree. What is it, the tree outside my window, a (unspecific one) tree, CS structure… Give a machine context and of it goes, no guessing necessary. I really start wondering about all these Keyword (Google, Microsoft is actually a little better) guessers. If text were as unstructured as people think they would produce something like Cuil. And no, structure is not keywords. We just can guess (on not) what a text is about from keywords if they are very specific and as long as the Human writer doesn’t play games with the text, like in the early years of the web.

      We will see that it gets better as soon as search engines know when to stop, instead of announcing 10t found results. That was cute in the early days of the web, does anybody care today?