I really enjoyed reading this post, as I think Adam makes some excellent points. Whevener stuff like this gets to the frontpage of hackernews I get really excited, and make sure I carefully read each and every comment it gets. The reason for my enthusiasm for dating-related discussion can be easily explained: six years ago I cofounded a datingsite in the Netherlands which takes a completely different, and hopefully better, approach to online dating. And so far, the (dutch) online daters seem to agree with us: over the years we've consistently received very positive reviews (by members and the media in general), in stark contrast with what is generally being written about online dating. Since we have the ambition to move the concept to the US this year, I'm going to go ahead and actually start joining the discussion here by giving some details into how we've managed to address some of the issues Adam talks about in his post.
Our site, in a nutshell, basically doesn't allow members to seek out other members and message them to see if there's mutual interest. It's only possible to talk to people you've been matched to by our system. In order to increase the chances of interaction between the matched members we artifically limit the frequency at which one receives new matches. Also, we give very little information about the match. The system has good reasons to introduce the members to each other, they will have to find out for themselves what those reasons might be. This approach helps mitigate the issue of 'Hypergamy', as the matching system ensures the match works both ways. Furthermore it prevents spamming (it's just not possible) and deincentivizes fakers.
We can only get away with not allowing people to search for themselves because we have a great matching system. Adam talked about eHarmony's love-science-PhD's and noted a statistical approach would probably work better. That's exactly what we thought, and put into practice. By combining the information we have of our members (answers to questions, attractiveness scores, etc) with a feedback loop that tells us which matches actually work out (or don't), we can constantly improve our algorithm (think machine learning) to make better matches. We actually have spend years optimizing the crap out of this.
Now, for us to be able to have a fighting chance in the US, we need to also somehow solve the fabled chicken and egg problem. We managed to pull it off in the Netherlands (word-of-mouth mostly, a big plus of actually having a good product), but will have to do more to succesfully launch in a foreign market. If anyone here has thoughts about this and is willing to share, I would be very grateful.