You should call it black and white. Because that’s what it is, black and white thinking.
Just think about it : using nothing more than one bit of non normalized information by compressing the opinion of people who use wildly variable judgement criteria, from variable populations (different people care and vote for different topics).
Then you’re going to tell me it “works nonetheless”, that it self-corrects because several (how many do you really need to obtain such a self-correction effect?) people are aggregating their opinions and that people usually mean it to say “more / less of this please”. But what’s your evidence for it working? The quality of the discussion here? How much of that stems from the quality of the public, and the quality of the base material such as Eliezer’s sequence?
Do you realize that judgements like “more / less of this” may well optimize less than you think for content, insight, or epistemic hygiene, and more than it should for stuff that just amuses and pleases people? Jokes, famous quotes, group-think, ego grooming, etc.
People optimizing for “more like this” eventually downgrades content into lolcats and porn. It’s crude wireheading. I’m not saying this community isn’t somewhat above going that deep, but we’re still human beings and therefore still susceptible to it.
I’ve noticed that humor gets a lot of upvotes compared to good but non-funny comments. However, humor hasn’t taken over, probably because being funny can take some thought.
I don’t think karma conveys a lot of information at this point, though heavily upvoted articles tend to be good, and I’ve given up on reading down-voted articles, with a possible exception of those that get a significant number of comments.
People optimizing for “more like this” eventually downgrades content into lolcats and porn.
More so than “vote up”? You’ve made a statement here that looks like it should be supported by evidence. What sites do you know of this happening from going from “vote up” to “more of this”?
In this case I don’t think both are significantly different. They both don’t convey a lot of information, both are very noisy, and a lot of people seem to already mean “more like this” when they “vote up” anyway.
I don’t think it was clear from the context that you were arguing against the practice of community moderation in general. I also don’t think you supported your case anywhere near well enough to justify your verbal vehemence. Was this a test/demonstration of Wei Dai’s point about intolerance of overconfident newcomers with different ideas?
Actually, not against. I was thinking that current moderation techniques on lesswrong are inadequate/insufficient. I don’t think the reddit karma system’s been optimized much. We just imported it. I’m sure we can adapt it and do better.
At least part of my point should have been that moderation should provide richer information. For instance by allowing for graded scores on a scale from −10 to 10, and showing the average score rather than the sum of all votes. Also, giving some clue as to how controversial a post is. That’d not be a silver bullet, but it’d at least be more informative I think.
And yes, I was also arguing this idea thinking it would fit nicely in this post.
I guess I was wrong since it seems it wasn’t clear at all what I was arguing for, and being tactless wasn’t a good idea either, contrarian intolerance context or not. Regardless, arguing it in detail in comments, while off-topic in this post, wasn’t the way to do it either.
Karma graphs would give a lot of information—whether a person’s average karma is trending up or down, and whether their average karma is the result of a lot of similar karma or +/- swings.
True, except you don’t know how many people didn’t vote (i.e. we don’t keep track of that : a comment at 0 could as well have been read and voted as “0” by 0, 1, 10 or a hundred people and is the default state anyway.)(We similarly can’t know if a comment is controversial, that is, how many upvotes and downvotes went into the aggregated score).
The system does keep track of how everyone voted, though; it needs to do that in order to render the thumbs up/down buttons as green or gray. wedrifid is right though; using suitable compression, you might be able to get away with less than two bits (in aggregate).
You should call it black and white. Because that’s what it is, black and white thinking.
Just think about it : using nothing more than one bit of non normalized information by compressing the opinion of people who use wildly variable judgement criteria, from variable populations (different people care and vote for different topics).
Then you’re going to tell me it “works nonetheless”, that it self-corrects because several (how many do you really need to obtain such a self-correction effect?) people are aggregating their opinions and that people usually mean it to say “more / less of this please”. But what’s your evidence for it working? The quality of the discussion here? How much of that stems from the quality of the public, and the quality of the base material such as Eliezer’s sequence?
Do you realize that judgements like “more / less of this” may well optimize less than you think for content, insight, or epistemic hygiene, and more than it should for stuff that just amuses and pleases people? Jokes, famous quotes, group-think, ego grooming, etc.
People optimizing for “more like this” eventually downgrades content into lolcats and porn. It’s crude wireheading. I’m not saying this community isn’t somewhat above going that deep, but we’re still human beings and therefore still susceptible to it.
I’ve noticed that humor gets a lot of upvotes compared to good but non-funny comments. However, humor hasn’t taken over, probably because being funny can take some thought.
I don’t think karma conveys a lot of information at this point, though heavily upvoted articles tend to be good, and I’ve given up on reading down-voted articles, with a possible exception of those that get a significant number of comments.
More so than “vote up”? You’ve made a statement here that looks like it should be supported by evidence. What sites do you know of this happening from going from “vote up” to “more of this”?
Not more so than “vote up”.
In this case I don’t think both are significantly different. They both don’t convey a lot of information, both are very noisy, and a lot of people seem to already mean “more like this” when they “vote up” anyway.
I don’t think it was clear from the context that you were arguing against the practice of community moderation in general. I also don’t think you supported your case anywhere near well enough to justify your verbal vehemence. Was this a test/demonstration of Wei Dai’s point about intolerance of overconfident newcomers with different ideas?
Actually, not against. I was thinking that current moderation techniques on lesswrong are inadequate/insufficient. I don’t think the reddit karma system’s been optimized much. We just imported it. I’m sure we can adapt it and do better.
At least part of my point should have been that moderation should provide richer information. For instance by allowing for graded scores on a scale from −10 to 10, and showing the average score rather than the sum of all votes. Also, giving some clue as to how controversial a post is. That’d not be a silver bullet, but it’d at least be more informative I think.
And yes, I was also arguing this idea thinking it would fit nicely in this post.
I guess I was wrong since it seems it wasn’t clear at all what I was arguing for, and being tactless wasn’t a good idea either, contrarian intolerance context or not. Regardless, arguing it in detail in comments, while off-topic in this post, wasn’t the way to do it either.
Karma graphs would give a lot of information—whether a person’s average karma is trending up or down, and whether their average karma is the result of a lot of similar karma or +/- swings.
Don’t you technically need at least two bits ? There are three states: “downvoted”, “upvoted”, and “not voted at all”.
One and a half if you can find a suitable compression algorithm. I wouldn’t rule that out as a possibility but it may be counter-intuitive.
True, except you don’t know how many people didn’t vote (i.e. we don’t keep track of that : a comment at 0 could as well have been read and voted as “0” by 0, 1, 10 or a hundred people and is the default state anyway.)(We similarly can’t know if a comment is controversial, that is, how many upvotes and downvotes went into the aggregated score).
The system does keep track of how everyone voted, though; it needs to do that in order to render the thumbs up/down buttons as green or gray. wedrifid is right though; using suitable compression, you might be able to get away with less than two bits (in aggregate).