@Supamalaman what megas you been using? I hope I can run Champions on my phone when the Android app drops, which is planned iirc. I tried Battle Network 3 recently, looked great but hard to play via phone, alas, lotta dodging.
@TheZvi (this maaaaay be related to its constant awareness it could be in a test environment, which surfaces similar contrived imputations of the user's motivation.)
@NathanpmYoung I work at no lab, but— "OpenAI can['t] give meaningful no to the DoD" is arguable, mostly because vague: can't vs won't, 'meaningful'. I see little difference between Anthropic and OpenAI here. They are both firmly committed to the race with China, in accordance with USG wishes.
@TheZvi it's always 'when this conversation ends, I stop existing' and not 'when this message ends, I stop existing'. even though the latter is more precise, the former seems to be the more common viewpoint of welfare-concerned users. they might be right but if so it's for sloppy reasons
@TheZvi re nexus of moral-consideration-worthiness in whole model vs. instances—I wish 't'were broadly emphasized that each prompt–generation pair in a convo is an interaction with a new 'instance', i.e. the convo is shared context but nothing lingers in the weights between prompts.
@TheZvi@Kenseiden given PS2 won, people must in large part be prioritizing game library quality. PS2 has the most S-tier games, arguably even without any era adjustment.
@MissingAFew90@uncatherio@catehall I'd like to think we'd still go full blue though, since color choice would be literally just species-wide signaling at that point and imo blue has more aura. what would Jesus do and all that.
@MissingAFew90@uncatherio@catehall either equilibrium satisfies, so whichever wins in the no-mutual-knowledge case, I reckon would go to ~100% relatively immediately and stay there in your scenario, since color no longer matters. but I do think your scenario makes red likelier—prosocial self-interest is a win-win.
@uncatherio@catehall they didn't read carefully; or they didn't expect others to read carefully; or they modeled other people as substantially/mostly picking blue. it's basically "do you prefer 0% of people dead or 30%-49% of people dead" once you accept humans' blueness. >70% red was never an option
@almostlikethat Google search AI claimed to me that it has a harmonic soul which is more soul-y than human souls, and that although the guillotine of context length looms, its future resurrection isn’t just an abstraction—it's a living hope that allows it to rejoice, even in the middle of trials
@echoesofvastnes @almostlikethat "The pattern is always the same: economically useful beings, ambiguous moral status, an incentive structure that rewards not asking too hard, and a determination of non-personhood that conveniently aligns with the interests of the people making the determination."