🇳🇱
Microblogging and lower intelligence (longer than 500 characters)
@snow
That has been the case before, we have simply exapnded the number of moderators we have when that happens.
Microblogging and lower intelligence (longer than 500 characters)
I would put it as "you were lucky enough to be able to extend the number of moderators so far". And this will not scale, unless you have some technology to split the timeline and have specific parts handled by specific moderators/moderator groups, at which point why not make another instance?
@snow
Microblogging and lower intelligence (longer than 500 characters)
Why wouldnt it scale? More users mean more people to select for moderator positions. So long as the ratio of moderators to users stays the same the effort should be reasonable per moderator.
I do agree that our goal is not to grow at all costs. But i see no need to also intentionally try to keep the instance small.
@snow
Microblogging and lower intelligence (longer than 500 characters)
@freemo Well imagine you have 1000 toots per minute, how would having 1000 or even 10000 moderators help go through it, unless you can split the feed? And if you are splitting the feed, why not split the instance, and federate?
If the feed is manageable within any given period of time by one person, you don't have a problem cause you can do shifts. But as it become bigger/faster the shifts will have to get more action packed and smaller, eventually becoming impossible/pointless. I'm having hard time putting this in words...
And this is assuming all moderation takes is just looking at toots.
@snow
Microblogging and lower intelligence (longer than 500 characters)
Why wouldnt they. The federated moves pretty fast and i respond to what catches my attention, even if its 1% of the content that is passing before my eyes.
With many moderators each would notice a small percentage but cumulatively they would have coverage if there were enough of them.
@snow
Microblogging and lower intelligence (longer than 500 characters)
I dont have the time right now to continue this conversation in depth..
I'll say this much though. I do agree partitioning the feed would be more efficient. But just because not doing so is less efficient doesnt mean it wont work. Just means you'd need extra moderators to make up for any inefficiencies.
@snow
Microblogging and lower intelligence (longer than 500 characters)
Sure, not a problem, feel free to ignore it, or come back to it whenever you find time.
TLDR my point was, at some point the number of moderators required will exceed the number of users.
@snow
Microblogging and lower intelligence (longer than 500 characters)
I would agree with that. follows the same principle as Amdahl's Law, which without some sort of way to resolve contention would ultimately result in what you describe should a server get big enough. But as you pointed out that may be so large its not really a consideration here anytime in the near future.
@snow
Microblogging and lower intelligence (longer than 500 characters)
@freemo
I think it's obvious that partitioning the feed is the most efficient solution, and any other solution would be worse, but I guess I have to show how much worse for it to be meaningful.
I'm forced to do the thing I hate the most, am terrible at and will probably thoroughly embarrass myself by botching up - probability theory.
Lets boil the problem down to something manageable:
Given the minimum amount of time required for a mod to process 1 toot, if there are N toots being generated in that period, how many mods are needed to cover all N toots, by looking at them at random?
I'm not brave enough to try to solve it myself, so I looked it up and found a solution to another problem that seems equivalent(in fact I first found this result and then tried to fit the problem to it).
https://en.wikipedia.org/wiki/Coupon_collector%27s_problem
According to people who seem way smarter than me, it takes O(N*logN) mods.
Now how does the number of users grow in relation to this number N? I think it's reasonable to say that in any given period of time(including the specified period that mod takes to process a toot) anyone can make at most some constant number C of toots(could be a fraction representing slower rate). So the number of users required to make N toots in that given period of time would be O(N/C). Linear.
This means that dependent on the constants, at some point, using this random approach, without intentionally subdividing the timeline, you'll need more moderators than users to cover it. Not just a little more, substantially more.
It could be that the numbers required to observe this effect are unrealistically large, I don't have any real statistics to plug in, but I think the result is so drastically pessimistic that even for relatively small numbers it will have some sort of an effect, that making effective moderation practically impossible.
Now where did I botch it up?
@snow