4 ms·
Paul seems to be hitting on the optimization by proxy issues seen in many areas, even in pagerank. Google looks not for quality of a page but for indicators of
by alexandros 18y ago
Paul seems to be hitting on the optimization by proxy issues seen in many areas, even in pagerank. Google looks not for quality of a page but for indicators of that quality, namely inbound links by quality websites. Similarly exams and credentials do the same. If you can write about these books, you know literature. If you got a degree there, you must be very capable. And in the first generation it works.
However, as people become aware of the ranking mechanism, they try to get ahead by overoptimizing for the narrow areas inspected. So pages do linkfarms and black-hat SEO, students goto cram schools, etc. etc. essentially producing the exact opposite of what was initially sought. I believe Joel spolsky had a good article on a simillar pattern in sales.
The question is if an algorithm that affects its input data can be made so that it adapts to the changes and gradually reaches some kind of equilibrium with the input data or stabilizes in some other way. As paul suggests, one way would be to make the system more transparent, or even 'social' such that the participants have a way to update it in an agreed way. This would keep the minority from cheating on the majority.
Doing away with such filters though might be a bit harder than we think, especially given the breadth of their applications. As aristus commented, even YCombinator has an examination system. In fact it wouldnt be too hard to imagine cram schools for YC given enough incentive. I fear this is a problem we have to either solve or live with.
- ryanwaggoner 18y agoI fear this is a problem we have to either solve or live with. Aren't they all? :-)
- netcan 18y agoHow would you run a cram school for YC? Say you were given 19 yr olds at the end of Year 1 of Uni. You have 2-3 years part time. How would you get them in? Activities? Interview coaching? Resume building? Portfolio building? Specialised learning?
- alexandros 18y agoIntresting question, but the feasibility of it is not dependent on my ability to give a good answer. Regardless, I'll give it a shot for argument's sake. Of course everything you say is valid. you would need to do these things. I would also mine all the successes, details of interview process and characteristics of successful applicants and their applications for commonalities and then see how many of these can be easily added to an application for 'bonus points'. Of course it depends on how black-hat you want to get about this. The first thing that came to my mind though, having read pg's essays, would be 'teach them LISP' :)
- netcan 18y agoMy intention wasn't to corner you, just to start a discussion. For arguments sake, let's get about as black hat as a Chinese mother with a shot at Oxford. I don't know much about YC or what influences them apart from fragments from pg's essays. Insisting that LISP or some other language no one uses is absolutely the best tool seems like it might be a bonus point. But I think that is as a bonus. What would be the core? They say smartness & an attitude that they are going to do something big with or without the program. The latter can be gamed. The former, I don't know much about the criteria.
- derefr 18y agoI'd probably build their social network: connect them to hundreds of influential sources of free marketing. If you're judging performance, the initial performance of a connected founder will greatly exceed an unconnected one. If you don't wait long enough to watch market effects take over before judging which startups to keep, you'll only end up with the networked ones.
- netcan 18y agoNot sure about gaming YC, but that might be a trick for gaming performance measurements.
- motoko 18y agoIf the leading criterion for admission was concrete evidence that the applicants collectively have the ability and motivation build things people want as measured in users and sales... ...then a cram school for YC would be a program much like YC in which younger teams built smaller, more easily guided technology start-ups.
- netcan 18y agoIf the leading criterion for admission was concrete evidence that the applicants collectively Big if. I'm sure that any self respecting University will tell you that the best prep for their University is anything that makes the candidates smarter, more driven & more likely to succeed. Everyone uses indicators. Some are better then others. Maybe some can never be gamed (I doubt it) but as actually mentioned in this essay, the way around credentials is making entry less important.
- motoko 18y agoI meant admission to YC, not university. But on that note, isn't the biggest criterion for university transfer admission is who admitted you in the first place and your consequential success at that school? It stands to reason if one wanted to predict if a group would be successful creating a successful startup that a good predictor would be if that group had stated a successful startup in the past. I'd assume there are other criteria, like: is what this group doing within our scope of expertise? Can we add value to this startup? Will admitting this group best add to the quality of this YC class? Will we make money on this investment?
- pg 18y agoThat's a very interesting question. Above all I'd have them build things. We like anyone who's created an open-source project we've heard of, for example. I'd also try to get them jobs at existing YC startups. Then they'd both learn how startups work, and (if they were good) get recommendations from people we know.
- netcan 18y agoIs that the equivalent of directly influencing them (cram school like?) or indirectly (actually preparing them?)