@scriabingroyper@nosilverv "prole overproduction" guys this is just eugenics it is verifiably not even a tiny bit new. do you need help looking up the sources?
@RuxandraTeslo I think here it's just that the argument is coming from a value of mercy ungrounded by the vicissitudes of applying a measurement apparatus to something like "need." You may forget just how far above your perspective is given your stats education.
@NandoDF I thought give the loss function change, we really do mask what action we took that yielded the observation, and so you could instead weaken its presence rather than fully remove it.
@NandoDF perhaps tho if the policy represents memory (accumulated previous observations in the form of the taken action), one may want a modulating term on weighing it as information rather than totally removing it?
@PatrickToulme Also no offense intended, but this article definitely pings my "Claude wrote this" radar. There's certain rhetorical flourishes and the dreaded high emdash budget that make me think this. Really appreciate the content but because I'm reading Claude all day I get a bit glazed over