I think it’s kind of inevitable that LessWrong will eventually get a huge number of people attempting to join. I do think we’ll need to deal with that somehow or other sooner or later. I don’t think we’re ready yet. I think it’s possible for us to become ready if we prioritize it.
We’ve been thinking about this a lot lately. In addition to the rejected section we also recently shipped AutoRateLimits for low and negative karma users. I’ll have a post about this soon, but the basic gist is that users start with a rate limit of 3 comments per day and 2 posts per week. That rate limit disappears once they hit 5 karma. If their karma becomes negative, this limit is made more strict. (At −1 karma, it becomes 1 post per week and 1 comment per day. At −15 karma they can only write 1 comment every 3 days. At −30 karma they can only submit 1 post every 2 weeks)
I don’t think this is enough to handle a really large influx. Even if they’re heavily rate limited, a bunch of people posting a mediocre comment every 3 days would add up to a significant drop in the site signal noise ratio. You could increase the rate limit but that does make things harder for new users and would make the site feel more punishing. Karma is only a rough measure of quality and there’s a lot of room for disagreement over whether a given downvote is fair.
In the pre-GPT world I’d be more optimistic about making some kind of test that checks for whether a user has a reasonable understanding of what LessWrong is about. and is able to participate. In the post GPT world it’s less clear how to do that sort of thing – any kind of automated test is basically a test for “do they know how to use an LLM?”.
There are options like “Let established users approve new users.” I’d set my bar fairly high for established users to avoid a situation where each generation of users lets in a somewhat weaker set of users. The kind of bar I’d feel safe with for users-with-approval-power is something like “They’ve gotten 2-3 posts highly uploaded in the LessWrong Annual Review”.
That all said, I’m not actually sure how big an influx we’re likely to be talking about from your video. Would it be larger than the TIME article?
You could have AutoAutoRateLimits. That is, you have some target such as “total number of posts/comments per time” or “total number of posts/comments per time from users with <15 karma” or something. From that, you automatically adjust the rate limits to keep the target below a global target level. Maybe you add complications, such as a floor to never let the rate limit go to None, and maybe you have inertia. (There’s plausibly bad effects to this though, IDK.)
Some rough thoughts:
I think it’s kind of inevitable that LessWrong will eventually get a huge number of people attempting to join. I do think we’ll need to deal with that somehow or other sooner or later. I don’t think we’re ready yet. I think it’s possible for us to become ready if we prioritize it.
We’ve been thinking about this a lot lately. In addition to the rejected section we also recently shipped AutoRateLimits for low and negative karma users. I’ll have a post about this soon, but the basic gist is that users start with a rate limit of 3 comments per day and 2 posts per week. That rate limit disappears once they hit 5 karma. If their karma becomes negative, this limit is made more strict. (At −1 karma, it becomes 1 post per week and 1 comment per day. At −15 karma they can only write 1 comment every 3 days. At −30 karma they can only submit 1 post every 2 weeks)
I don’t think this is enough to handle a really large influx. Even if they’re heavily rate limited, a bunch of people posting a mediocre comment every 3 days would add up to a significant drop in the site signal noise ratio. You could increase the rate limit but that does make things harder for new users and would make the site feel more punishing. Karma is only a rough measure of quality and there’s a lot of room for disagreement over whether a given downvote is fair.
In the pre-GPT world I’d be more optimistic about making some kind of test that checks for whether a user has a reasonable understanding of what LessWrong is about. and is able to participate. In the post GPT world it’s less clear how to do that sort of thing – any kind of automated test is basically a test for “do they know how to use an LLM?”.
There are options like “Let established users approve new users.” I’d set my bar fairly high for established users to avoid a situation where each generation of users lets in a somewhat weaker set of users. The kind of bar I’d feel safe with for users-with-approval-power is something like “They’ve gotten 2-3 posts highly uploaded in the LessWrong Annual Review”.
That all said, I’m not actually sure how big an influx we’re likely to be talking about from your video. Would it be larger than the TIME article?
You could have AutoAutoRateLimits. That is, you have some target such as “total number of posts/comments per time” or “total number of posts/comments per time from users with <15 karma” or something. From that, you automatically adjust the rate limits to keep the target below a global target level. Maybe you add complications, such as a floor to never let the rate limit go to None, and maybe you have inertia. (There’s plausibly bad effects to this though, IDK.)