I’m not sure, but keep in mind that karma score isn’t exactly ‘how good a post is’, it’s a proxy—and intelligences optimizing for a proxy measure can end up stomping on your supergoals.
Edit: There’s a new utility-function proposal—have your AI simulate each member of LW, and have them up/downvote its ideas!
Is content designed more specifically to be liked by others really a problem?
I’m not sure, but keep in mind that karma score isn’t exactly ‘how good a post is’, it’s a proxy—and intelligences optimizing for a proxy measure can end up stomping on your supergoals.
Edit: There’s a new utility-function proposal—have your AI simulate each member of LW, and have them up/downvote its ideas!