Also on my project list - the AntiAssholeFilter. It's so interesting the ways you could handle this. Personally, I would want to just transform the comment into something that doesn't even mention that that the commenter wrote a mean comment - if it has something of value just make it not mean, otherwise hide it.

A couple things are really interesting about this idea. First - it's so easy for the end-user to customize a prompt that you don't need to get it right, you just need to give people the scaffolding and then they can color their internet bubble whatever color they want.

Second, I think that just making all comments just a couple percent more empathetic could be really impactful. It's the sort of systemic nudge that can ripple very far.