Re:How do you reason about a probabilistic distributed system?
by Swizec on Friday September 11, 2020 @07:36PM
(#24448468)
My favorite version of this is that shuffle mode is really hard to implement for music . You can’t use random sort because it leads to too many repetitions of songs close together. When that happens people complain that your shuffle isn’t random enough.
Re:How do you reason about a probabilistic distributed system?
by colinmhayes on Friday September 11, 2020 @10:50PM
(#24449403)
This, but spotify's shuffle is actually horrendous. It only plays songs within something like 50 of the one it's on. If you have a multi-hundred playlist you'll only hear the same 50 songs if you put it on shuffle and click the first song until it's finished with all 50 of those songs. Then it'll go to the next 50.
When something grabs your attention, you start to suffer from Frequency Bias (like 'all of a sudden' everyone has the same car you bought or are going to buy). But in this case, everyone really does suck at statistics. Which probably explains why when I learned what an actuary was and that they made about the same as an anesthesiologist, I wasn't that surprised.
The ironic part for me is that I kept thinking dependent variables were independent, whereas most of the people I observe think independent variables are dependent (gambler's fallacy, etc), and so vastly underestimate the potential for a 'bad run' of events.
When I still played WoW they posted an article about how they tweaked probabilities a bit to avoid frustrating people. Much of the questing in WoW is based on succeeding at an action a set number of times that has a probability of failure, which turns it into a normal distribution. Your buddy might have gotten lucky and finished a quest in two minutes. Meanwhile, there's a 1 in a million chance that you could still be working on that quest 5 hours later. And there are 10 million users...
So to prevent rage quitting, they made the success rate dependent. Which they can do because the scenario is fictitious. You can't do the same thing in the real world.
I'm not sure if there's a way through this or around it (except maybe "always hire a professional"). More likely in 10 years we will discuss how we should have just backed away slowly from these probabilistic situations and done something else.