Yeah it's spammy, but surprisingly effective, I'm encouraging him to write an A/B testing module for it so he can test the effectiveness of different messages.
Contrary to what one might think he's just using this for LTRs.
I agree that there are people who prefer the time-saving methods used with many of the major online dating services, and they will certainly continue to use those, but we feel they may be missing out on people that they didn't expect.
Instead of matching profiles, match messages, people looking for long term populate one set of bayesian filters, and people looking for short term populate another set of bayesian filters. The spam cut and paste message go away pretty quickly.
The mechanism could be as simple as after a message is read they are prompted as to whether they liked the message or didn't like it, combine with a feature sorting messages by their bayes score?
Displaying your own ads might be something to think about - collecting user info over time and delivering highly targeted ads might be more profitable.
I don't necessarily disagree with most of what else you say in the blog post, but if this is seriously your conception of how people use dating sites, then you are doomed to fail.
Speaking from the experience of living with various roommates who have done e-dating with the popular services, none of them have ever approached them in this mechanical way. The data helps refine the search, but they're still looking for attractive people to chat with.
And that data part is important. If you're a 20-year-old looking to have fun, you're going to want to weed out anyone who is in an age demographic that wants to have a life-long relationship. And vice versa. So not having that data is a disadvantage that you have to make up for.