I've heard people describe a situation in ML (could be hill climbing) where an optimising algorithm will get stuck in a valley. At that point it will try various things in order to get unstuck, often involving a good dose of randomness. What's to say this didn't happen to op or that you really won't start seeing recommendations for Indian train death videos?
> What's to say this didn't happen to op or that you really won't start seeing recommendations for Indian train death videos?
We can never be sure of anything. What if HN start pushing train death videos on the front page and the admin can't tweak the algo because a glitch in the stack prevent them from accessing the backend ?
