Re:“Lifefaker.com makes faking perfection easy”
by siavosh on Sunday April 29, 2018 @11:38PM
(#16955544)
I think that’s only true if it’s a predictable reward function; if you add a random variable, in the words of Mr. Miyagi, “no can defense.”
[ Parent
| on HN ]
Re:“Lifefaker.com makes faking perfection easy”
by oblio on Monday April 30, 2018 @03:46AM
(#16956607)
https://www.joelonsoftware.com/2018/01/12/birdcage-liners/
> How do we prevent extinction? By only giving treats some of the time. So the dog learns something more subtle. When my master says come and I obey, I might get a treat. Sometimes I do, sometimes I don’t. That way, if I obey and don’t get the treat, I shouldn’t panic. I should still always come when he says come because that’s still the best way to get the most treats. Intermittent reinforcement works better.
> This sounds like what Facebook was doing to me.