2 ms·

First of all, article is incorrect: for a long time now YT optimizes recommendations not for watch time, but for life-time engagement. YT probably has the most

by vl 6y ago

First of all, article is incorrect: for a long time now YT optimizes recommendations not for watch time, but for life-time engagement. YT probably has the most advanced QT ML in the world. Second, it’s trivial to re-adjust recommendations: go to history and delete everything you didn’t like, and recommendations instantly will become much better.

disgruntledphd2 6y ago
> First of all, article is incorrect: for a long time now YT optimizes recommendations not for watch time, but for life-time engagement. YT probably has the most advanced QT ML in the world. Lifetime engagement is a function of watch time, right?
- vl 6y ago
  No, it turns that if you optimize for immediate watch time, people watch a lot at once and then didn’t return. It’s more profitable if people watch a little bit everyday for the rest of their life.