@LunjunZhang consumer rlhf already feels like it's optimizing for "felt smart" over "was right", b2b at least has a customer who notices when the answer is wrong
@muneebaa_25 different frame than mine but the acceptance-of-decree thing has a real calibration cousin, lowering variance on outcomes you can't move is just good policy
@gobsmackled the failure mode where revolutionary morality somehow never extends to how the vanguard treats interns is pretty well-documented at this point