I find myself increasingly in favor of tech solutions to moderation problems. It can be hard for users to change their behavior based on a warning, but perhaps you can do better than just ban them – instead shape their incentives and save them from their own worst impulses.
Only recently has the team been playing with rate limits as alternative to bans that can be used to strongly encourage users to improve their content (if by no other mechanism to incentivize investing more time into fewer posts and comments). I don’t think it should be overly hard to detect nascent Demon Threads and then intervene. Slowing them down both gives the participants times to reflect more and handle emotions that are coming up, and more time for the mod team to react.
In general, I’d like to build better tools for noticing places that would benefit from intervention, and have more ready trigger-action plans for making them go better. In this recent case, we were aware of the exchanges but didn’t have a go-to thing to do. Some of this was not being sure of policies regarding certain behaviors and hashing those out is much slower than the thread proceeds. In my ideal world, we’re more clear on policy, we know what our tools are, so it’s easy to act.
It might be apparent to everyone, but late 2021, day-to-day leadership went from Habryka to me as Habryka went to lead the newly created Lightcone Infrastructure more broadly. My views on moderation are extremely downstream of Oli’s, but they’re my own, and it’s taken time for me to have more confident takes on how to do things (Oli’s views are not so well codified that it would have even been feasible to just try to do what he woudl have done, even if I’d want to do). All that is a long way of saying that I and the current team are to some degree building up fresh our moderation policies, and trying to build them for LessWrong in 2023 which is a different situation than in 2018. I hope that as we figure it out more and more, it’s easier/cheaper/faster for us to moderate generally.
Technological Solutions
I find myself increasingly in favor of tech solutions to moderation problems. It can be hard for users to change their behavior based on a warning, but perhaps you can do better than just ban them – instead shape their incentives and save them from their own worst impulses.
Only recently has the team been playing with rate limits as alternative to bans that can be used to strongly encourage users to improve their content (if by no other mechanism to incentivize investing more time into fewer posts and comments). I don’t think it should be overly hard to detect nascent Demon Threads and then intervene. Slowing them down both gives the participants times to reflect more and handle emotions that are coming up, and more time for the mod team to react.
In general, I’d like to build better tools for noticing places that would benefit from intervention, and have more ready trigger-action plans for making them go better. In this recent case, we were aware of the exchanges but didn’t have a go-to thing to do. Some of this was not being sure of policies regarding certain behaviors and hashing those out is much slower than the thread proceeds. In my ideal world, we’re more clear on policy, we know what our tools are, so it’s easy to act.
It might be apparent to everyone, but late 2021, day-to-day leadership went from Habryka to me as Habryka went to lead the newly created Lightcone Infrastructure more broadly. My views on moderation are extremely downstream of Oli’s, but they’re my own, and it’s taken time for me to have more confident takes on how to do things (Oli’s views are not so well codified that it would have even been feasible to just try to do what he woudl have done, even if I’d want to do). All that is a long way of saying that I and the current team are to some degree building up fresh our moderation policies, and trying to build them for LessWrong in 2023 which is a different situation than in 2018. I hope that as we figure it out more and more, it’s easier/cheaper/faster for us to moderate generally.
I might write more in a bit, will post this