What makes the problem more complicated still is, founders can change. E.g. when we meet someone who seems smart and energetic but a bit timid about starting a company, we have to ask ourselves if the guy's intrinsically timid, or just put off by the weirdness of the idea, like we were before we started Viaweb.

1. take every email exchange you've had with past YC companies whose outcomes are now known. Good=bought, IPO, etc., Bad=no longer exists (with exceptions for folks like Kiko?)

2. train a bayesian filter on the set

3. Apply to the rest with as yet unknown outcomes

4. Publish the results :D

Which makes me think have they run the software on their own company? But it makes for interesting reading. A quick look around it appears to be positioning itself as doing all things to all parties:

- dating site for startups & engineers

- business reporting (analysts) for VC's

- euro-style-facebook for engineers & entrepreneurs

"... We provide business competition, event, networking and mailing list functionality for real-world groups that support startup communities. For investors, we're building technology to help inform their decisions. ..." [0]

Interesting to see Harj [1] there (with a low id) and ycombinator listed as a Startup but not a direct competitor. [1] Is uNoodle trying to do too much at once? There is another problem with this approach. Startups make new things using Ideas so new that historical data points are no real predictors of success. So how predict such Startups are going to succeed?

Make no mistake, though. The real target for these guys is to set themselves up as middle men for VCs. Mini-blood suckers.

[0] http://younoodle.com/about.php

[1] Not very surprising due to the number of Oxford & Entrepreneur alumni ~ http://younoodle.com/profile.php?id=23

[2] http://younoodle.com/group.php?id=74

It is a paradox only if you assume that the quality of the product makes or breaks a company (not to mention their "AI" could be right only accidentally, in this particular case). It is weird that they believe that, given that "their algorithm uses sophisticated modeling pertaining to how social capital and networks can affect an organization’s performance".

It's an entertaining idea, but I doubt how they got to score such a PR hit with no substance whatsoever.

Hmm. I see a world where VCs and other investors use "the box" to pick the companies they invest in. I also see a world where startups use "the box" to adjust their companies "box score" to present to investors. In the end, "the box" may be a self-fulfilling AI system.

Coders: don't be fleeced of your valuable skills. If these people had good connections, foresight, understood technology or valued programming then they wouldn't "Need someone with coding skills" ( http://www.younoodle.com/profile.php?id=989 ), be "on the lookout for web design jobs" ( http://www.younoodle.com/profile.php?id=16 ) or "have leet skillz" ( http://www.younoodle.com/profile.php?id=4 ). That's just the founders' alumni ( http://www.younoodle.com/group.php?id=2 ).

the second guy is not "on the lookout for web design jobs", he wants to find people that are cuz he has too many people asking him if he knows anyone that can do it.

the third guy was joking, i assume. he co-founded the site, after all..

Do these guys fall under the 'entrepreneurs, who lack programming ability who are seeking..."?

Ridiculous. One of the silliest things I've ever read in NYTimes. Kedrosky was being way too kind on them.

As for their algorithm, it's probably along the lines of how connected you are on their site.

How did they score funding from Max Levchin? One of the cofounders worked at Slide. http://younoodle.com/about.php So I doubt it's due to their amazing AI alone.

It's just a social network, folks. You can make your own on Ning or Facebook.

A better predictor would probably be someone's comments here on N.YC.

Don't you mean, someone's karma, nickb? ;-)

Predicting the success of a start up is probably just as hard as predicting where the market is going to go tomorrow (if not harder). There are too many variables in play. I think this is just a PR stunt; it definitely created enough buzz!

Activity wise, there certainly are enough data to pull a reasonable sample size to create a model. However, the tricky part is to identify the right variables. In my opinion, unlike the stock market, the variables to start-up success is much difficult to quantify. Would love to see their theory behind their model though!

Then why should anyone else?

I have a hard time believing that anyone that would build such a thing could resist running it on themselves.

I don't know these guys, but if it were me:

Run software on self. It says I'm going to fail. There's a bug in the software! Tweak, tweak. Run software on self. It says I'm going to fail. There's a bug in the software! Tweak, tweak...

doesn't everyone know failure feed success? So them problem is more how to increase your chance of success, not tell you Yes, No if you will succeed

With what data are they building their models?

for future reference: bugmenot.com