I upvoted you (+1), but the point of my comment was to make sure that someone besides the author of a comment reads it so that the score more closely reflects its real value. If you do it recursively then we will have two kinds of comments to consider: replies to my comment and to your, so it will become harder to make comparisons.
I guess I should have said something like “and if you up/downvote a reply in this comment thread, please read and score all of its siblings accordingly”, since if it’s just me upvoting the problem will persist. I can only give one of 5 possible scores, they will be just a noisy estimate of the “ideal score” and older comments will have bigger magnitudes.
And even if that was fixed, it seems people are mostly ignoring what I said and commenting elsewhere. Oh well, at least I tried.
When I first wrote the comment, it seemed like the best possible strategy according to what I care about. It goodharts the LW karma system so that it loses correlation with intellectual progress or whatever else it is supposed to reflect. And yet it isn’t a pure “I will unfairly upvote you and you unfairly upvote me” like aphyer’s and Tofly’s comments, nor is it nonsense or bot-like behaviour, like what rank-biseral and maybe goodestheart are doing. Now I’m not so sure.
To the Lightcone team, if you want to make sure that scores actually reflect the quality of comments and that users are shown the best possible ones, you should probably implement a multi-armed bandit algorithm as the default sorting method, divide the scores by the number of impressions and then multiply by some constant (10 times the standard deviation?) so that they aren’t tiny. For comments with very few votes you could use some ML model as a prior or show the uncertainty to the user.
Oh, and another bad faith strategy consists on writing many ok comments instead of just a few good ones, the Good Heart Tokens a user receives aren’t actually proportional to the total value added. Please keep the score of this one at 1.
I upvoted you (+1), but the point of my comment was to make sure that someone besides the author of a comment reads it so that the score more closely reflects its real value. If you do it recursively then we will have two kinds of comments to consider: replies to my comment and to your, so it will become harder to make comparisons.
I guess I should have said something like “and if you up/downvote a reply in this comment thread, please read and score all of its siblings accordingly”, since if it’s just me upvoting the problem will persist. I can only give one of 5 possible scores, they will be just a noisy estimate of the “ideal score” and older comments will have bigger magnitudes.
And even if that was fixed, it seems people are mostly ignoring what I said and commenting elsewhere. Oh well, at least I tried.
When I first wrote the comment, it seemed like the best possible strategy according to what I care about. It goodharts the LW karma system so that it loses correlation with intellectual progress or whatever else it is supposed to reflect. And yet it isn’t a pure “I will unfairly upvote you and you unfairly upvote me” like aphyer’s and Tofly’s comments, nor is it nonsense or bot-like behaviour, like what rank-biseral and maybe goodestheart are doing. Now I’m not so sure.
To the Lightcone team, if you want to make sure that scores actually reflect the quality of comments and that users are shown the best possible ones, you should probably implement a multi-armed bandit algorithm as the default sorting method, divide the scores by the number of impressions and then multiply by some constant (10 times the standard deviation?) so that they aren’t tiny. For comments with very few votes you could use some ML model as a prior or show the uncertainty to the user.
Oh, and another bad faith strategy consists on writing many ok comments instead of just a few good ones, the Good Heart Tokens a user receives aren’t actually proportional to the total value added. Please keep the score of this one at 1.