I appreciate the need for dealing with low-quality AI-generated content, but I worry about the risk of intentional manipulation, and of near-banning those with unpopular views, here. Some possible ways to mitigate that:
Total Downvote Percentage (or Count) versus Karma
Whatever the merits of strongvotes in other contexts, a minority view that someone’s contributions are negative should not carry the day against a greater number of expressed votes and/or voters. Indeed, I would argue that ~2/3 of the votes should be negative before one could appropriately infer a community consensus that the user is detracting from the Forum. Strong downvoting by a committed group is the most obvious way to manipulate the system into silencing those with whom you disagree.
On the other hand, I am okay with requiring both specified negative karma and a specified negative downvote percentage/count, because I would be hesitant to apply a rate limit when a significant minority strongly believed that the user’s activity was valuable.
Requiring Net Negative Reaction Across a Number of Posts/Comments
Otherwise decent users occasionally write a pretty bad comment that gets seriously downvoted. In my view, rate limits are appropriate when the user has a demonstrated pattern of low-quality activity (or for new users, when the information that is available suggests low quality). As written, I think there is a risk that one or two bad comments out of the last 20 have a disproportionate influence on the rate-limit decision. I’m sure people here can come up with a better weighting formula than I could, though.
Manual Review
Lower-karma users should have access to a prompt appellate process, under which a moderator manually determines whether a rate limit is warranted. The messages which convey the rate limit should inform users of their right to appeal. The appellate moderator should not presume that the automatic action was correct.
For mid+-karma users, I would prefer that the algorithm not apply automatically, but instead trigger a manual review.
Why the difference? It would be unexpected for a mid+-karma user to suddenly “post a lot of lower-quality AI-generated content.” As a result, the risk of a given user being censored for unpopular views is higher when the user’s karma is mid+-level. People who have a history of positive contributions deserve more process, and the downsides of their erroneous exclusion are likely more significant. Finally, at least in the absence of strategic downvoting to trip the rate limits, mid+-karma users should only rarely trigger the limits, so reviews shouldn’t take much moderator time.
All that being said, I recognize there are limits to what can be done without investing too much more time into the existing codebase.
Thanks for sharing your thoughts Jason! We’ll take this into account when we review the rate limits, after we have a bit more data on its effects in practice.
I appreciate the need for dealing with low-quality AI-generated content, but I worry about the risk of intentional manipulation, and of near-banning those with unpopular views, here. Some possible ways to mitigate that:
Total Downvote Percentage (or Count) versus Karma
Whatever the merits of strongvotes in other contexts, a minority view that someone’s contributions are negative should not carry the day against a greater number of expressed votes and/or voters. Indeed, I would argue that ~2/3 of the votes should be negative before one could appropriately infer a community consensus that the user is detracting from the Forum. Strong downvoting by a committed group is the most obvious way to manipulate the system into silencing those with whom you disagree.
On the other hand, I am okay with requiring both specified negative karma and a specified negative downvote percentage/count, because I would be hesitant to apply a rate limit when a significant minority strongly believed that the user’s activity was valuable.
Requiring Net Negative Reaction Across a Number of Posts/Comments
Otherwise decent users occasionally write a pretty bad comment that gets seriously downvoted. In my view, rate limits are appropriate when the user has a demonstrated pattern of low-quality activity (or for new users, when the information that is available suggests low quality). As written, I think there is a risk that one or two bad comments out of the last 20 have a disproportionate influence on the rate-limit decision. I’m sure people here can come up with a better weighting formula than I could, though.
Manual Review
Lower-karma users should have access to a prompt appellate process, under which a moderator manually determines whether a rate limit is warranted. The messages which convey the rate limit should inform users of their right to appeal. The appellate moderator should not presume that the automatic action was correct.
For mid+-karma users, I would prefer that the algorithm not apply automatically, but instead trigger a manual review.
Why the difference? It would be unexpected for a mid+-karma user to suddenly “post a lot of lower-quality AI-generated content.” As a result, the risk of a given user being censored for unpopular views is higher when the user’s karma is mid+-level. People who have a history of positive contributions deserve more process, and the downsides of their erroneous exclusion are likely more significant. Finally, at least in the absence of strategic downvoting to trip the rate limits, mid+-karma users should only rarely trigger the limits, so reviews shouldn’t take much moderator time.
All that being said, I recognize there are limits to what can be done without investing too much more time into the existing codebase.
Thanks for sharing your thoughts Jason! We’ll take this into account when we review the rate limits, after we have a bit more data on its effects in practice.