One of my optimization criteria is discreditable-ness which I guess is sort of like optimizing for downvotes insofar as my audience really cares about credibility. When it comes to motivational dynamics there tends to be a lot of crossing between meta-levels and it’s hard to tell what models are actually very good predictors. You can approximately model the comment you replied to by saying I was optimizing for downvotes, but that model wouldn’t remain accurate if e.g. suddenly Less Wrong suddenly started accepting 4chan-speak. That’s obviously unlikely but the point is that a surface-level model like that doesn’t much help you understand why I say what I say. Not that you should want to understand that.
One of my optimization criteria is discreditable-ness which I guess is sort of like optimizing for downvotes insofar as my audience really cares about credibility. When it comes to motivational dynamics there tends to be a lot of crossing between meta-levels and it’s hard to tell what models are actually very good predictors. You can approximately model the comment you replied to by saying I was optimizing for downvotes, but that model wouldn’t remain accurate if e.g. suddenly Less Wrong suddenly started accepting 4chan-speak. That’s obviously unlikely but the point is that a surface-level model like that doesn’t much help you understand why I say what I say. Not that you should want to understand that.