I'm impressed. It's not just a fluff piece about how amazing it is having dropouts save the world.
Devin keeps trying to figure out why omnipresent ego inflation is a good thing. He hits all the major points: it's about mindshare, addiction, and manipulating people to get a higher CPM.
Is that all social media is doing? Playing psychological video games in ways
that form habits and drive revenue for Internet companies?
He takes interviewee's quips (reduce friction! share yourself!), actually thinks (instead of sitting dumbfounded when someone young talks about technology and business), then figures out why the fads of today are shallow substitutes for sincere social interaction.
One of the founders of a YC company called 1000Memories.com (it's FB for dead people, only more interesting) says he heard FB can already tell when you're about to break up with someone: certain communication patterns emerge.
Yes, I know that the amount of data people share on Facebook is staggering, but, you've got to be kidding me!
I totally wouldn't put it past them, with the proviso that the proper claim is that they are able to predict when someone goes from "in a relationship" on FB to "single" or "in a relationship [with someone else" on FB, and this isn't perfect but it is much better than you'd expect from random chance.
For example, I think that Facebook could identity a pool of 100,000 users tomorrow and predict a breakup within the next two weeks, and be right for over 20% of them. That is statistically improbable unless you expect relationships to average about 10 weeks. (Yeah yeah, handwavy math there. Wave wave.)
How you'd do this? The same way you do any AI problem: cheat like a mofo until it looks like a problem we've already solved. For example, come up with a feature space of, say, 100 things which are easily algorithmically checkable: messaged boyfriend in past day, messaged boyfriend more than 10 times in last day, messaged boyfriend in last week, messaged boy other than boyfriend in last day, ..., changed sexual orientation listed in FB, changed residence listed in FB, ... went off to college, whatever. Assign each factor a random number from 1 to 100.
Now, construct a Hadoop cluster, grab your favorite million Facebook accounts, and use any AI technique you like (simulated annealing, GA, whatever) to play with those scores until you find a decent fit where the dot product of that vector and a vector of binary tests against a user delivers high scores for people who break up and low scores for people who don't.
Then, grab a hundred million Facebookers, take the dot product for each of them against your best performing vector, and print out the 100k with the highest scores.
This will work. Indeed, it almost can't not work, if any of your constituents of the feature vector captured any useful information. (P.S. The experimental design here is ridiculously biased in favor of success. Real life sometimes is -- it is certainly true for FICO, for example, since success is "beats a human underwriter" and human underwriters suck, what with their sucky salaries and sucky accuracy and sucky inability to process a hundred thousand applications an hour.)
I just finished my internship for this semester. I had the opportunity to see how death cases are investigated.
One case in particular involved cell phone records. After transcribing all of the calls into a ordered list a clear pattern emerged. When a crime is about to be committed, phones ring off the hook. This was confirmed by a state agent later.
The idea that FaceBook can predict patterns in people's relationships with others is not all that surprising; especially since they have context. Language is a good predictor of many things. Check out 'Gender Genie', a free program to predict the sex of an author.
Nah, I'm pretty sure that's just a Chinese Whisper that originated from David McCandless work in data mining breakups on Facebook. It can tell you what the most popular day to breakup is, but it can't predict when a couple is going to breakup.
However that said it probably is possible to predict breakups with communication data. You could probably even do it with an iphone app, breakups tend to follow certain events (like drop in respect, distancing, etc.) that can detected in vocal patterns.
There is a (YC?) company that analyzes your GMail and tells you which of your friends you haven't been in touch with, in recent times. I'm forgetting the name right now. But we all know that Facebook has an edge over GMail in terms of personal data, and they can certainly do it.
Hmm. I can see an app waiting to be built there. :-)
"Can tell", surely, with some accuracy better than random but not nearly good enough for actual predictions.
It would be interesting to know how the algorithm works, though... if only for my own personal usage. Is it a flurry of sudden communication? Or one partner starts ignoring the other? Do you need to know the words they're using, or can you just get it from the frequency of various types of communication?
Facebook does scare me enough, however, that I refuse to be friends (on facebook or in real life) with anyone who works at facebook. When I meet someone socially who works at facebook (which has only happened a couple of times so far) my next reaction is to edge slowly away from the conversation.
And you'd have to be crazy to date someone who works at facebook. And now I come to think of it, all the facebook employees I've met have been single...
I found his observations eloquent and for the most part very insightful. If nothing else, he does an excellent job of glamorizing why this market is so exciting:
"These people are optimistic not only because theirs is the last ascendant American industry but because implied in all those products is the idea that the human problem can be solved."
Edit: I just took a look at a random sample of startups in the batch we just accepted. Out of 21 founders, 2 went to Stanford, and 2 went to Ivy League colleges.
Generally, the meaning of orthogonal in conversation has become orthogonal to the "real" meaning: It can mean both 'perpendicular' as well as 'parallel.' "At odds" and "unrelated." It's really kind of silly.
If one of YC's methods is to keep up a certain level of excitement, and that's tough to do if you're building something highly functional but with little coolness value. Arguably, this correlates with degree of success[1]. Today it's social apps, tomorrow, it's Web 4.0.
It also seems apropos, considering that YC is, itself, a social program. I've only been to one of their semi-public events, as well as a couple loosely affiliated ones, and they were all clearly biased toward extroversion.
Similarly, I'd suggest "doomed" is hyperbole, as well, unless you mean getting accepted into their program. As a someone heavily introverted[2], it might be a form of torture, resulting in its own doom.
[1] Assuming success instead of failure. I agree there's something to be said for the potential scale of social apps compared to business tools, We haven't yet seen the likes of Facebook, and Zynga operate as public companies, without venture money pumping into the ecosystem. This tends not to be an issue, except with tools for startups.
My favourite quote from the article: "a guy named Rahul, a slender 27-year-old Englishman of Indian extraction who's 50 percent hair and 50 percent brain." Which is a hilarious description of YC/HN's own http://news.ycombinator.com/user?id=rahulvohra
Devin keeps trying to figure out why omnipresent ego inflation is a good thing. He hits all the major points: it's about mindshare, addiction, and manipulating people to get a higher CPM.
He takes interviewee's quips (reduce friction! share yourself!), actually thinks (instead of sitting dumbfounded when someone young talks about technology and business), then figures out why the fads of today are shallow substitutes for sincere social interaction.