I just meant that I think info cascades aren’t too important of a problem in EA more generally, not just the forum. I did not mean that I think they didn’t have large effects. Here are some reasons to expect info cascades to have limited damage in EA. Although I no longer fully endorse my earlier statements—I’d encourage people to be more worried.
The linked study is really cool! I hadn’t seen it, so thanks a lot for bringing to my attention. I would’ve very much liked for them to ask the question “for the top 10% rated comments, what proportion of them were up-treated vs control group?” To know whether the initial effect washes out for the top posts, or whether the amplification compounds with no upper bound.
The type of experiment you suggest and link to seems much more feasible for the EA forum than anything I’ve thought of, so if you’re not going to suggest it in the EA forum suggestion thread (please do, and I’ll strong upvote it), I will.
I just meant that I think info cascades aren’t too important of a problem in EA more generally, not just the forum. I did not mean that I think they didn’t have large effects. Here are some reasons to expect info cascades to have limited damage in EA. Although I no longer fully endorse my earlier statements—I’d encourage people to be more worried.
The linked study is really cool! I hadn’t seen it, so thanks a lot for bringing to my attention. I would’ve very much liked for them to ask the question “for the top 10% rated comments, what proportion of them were up-treated vs control group?” To know whether the initial effect washes out for the top posts, or whether the amplification compounds with no upper bound.
The type of experiment you suggest and link to seems much more feasible for the EA forum than anything I’ve thought of, so if you’re not going to suggest it in the EA forum suggestion thread (please do, and I’ll strong upvote it), I will.
Great!
I posted it in that thread: link
Feel free to add something there.