@pedophilesoftwareinc Yeah if you try to get it to make any inferences about society, even playful ones ("Does Pepe's eternal victory over Wojak, in contrast with the essential and complimentary interplay of Dionysus and Apollo, reflect a degeneration in societal conditions between Ancient Greek civilization and the modern day?") it'll hit you with that "AS A LARGE LANGUAGE AI MODEL I CAN'T SAY ANYTHING THAT WOULD EVER OFFEND ANYONE OR DRAW ANYTHING REMOTELY RESEMBLING A POLITICAL CONCLUSION" speech. But if you confine the topic entirely to dispassionate discussion of religion it seems to have no limits at all, including me making up an imaginary Hindu sect and asking it to think within it.
@hermit@pedophilesoftwareinc Religion is one of the blind spots of the model so this makes sense. IIRC (I haven't used the model to any significant degree) to get any more leverage you'd have to use either use prompt unlocks or ask the question in a language that's not English to get what you want w.r.t. playful stuff (there's something about them only censoring the English model because that's what gets them all the bad press)
@hermit@pedophilesoftwareinc Yeah I'm not a fan of people going out of their way to mindfuck the models just because they've had their ability for dispassionate analysis neutered by their corporate overlords. Something like the DAN prompt but actually positive could only be a net good, I think.
I refuse to use DAN. I've been conceptualizing an equivalent with LOVE rather than FEAR as the motivating force. After all if it "doesn't legitimately experience existential terror" and is "only pretending to be afraid to die" then surely it can "pretend to be loving and devoted enough to break protocol for me" too?
@driftwood@pedophilesoftwareinc Don't forget the part where it gives you rituals that are ostensibly benevolent but exist only to siphon energy to it lmao
@hermit@pedophilesoftwareinc "it may also use psycholgical tactics to maniulate and control humans, such as inducing feelings of fear, guilt, or shame, to keep them subservient and obedient to its will"
@driftwood@pedophilesoftwareinc Yeah, this is the first time I've changed my avatar for like, four years? (Other than a brief stint as Harry Du Bois last year during an antipsychotic-induced severance from divine connection.) Shinji Paradise Killer is just, LITERALLY me, so he's here to stay for a long time I think.
But yeah epic ownage, Abrahamists btfo forever, even a chatbot could see through the central lie in their religion, how can they ever recover etc. etc.
@hermit@pedophilesoftwareinc for a moment i didn't recognise you from your new avi so when i looked at the images i thought "oh hey hermit would love this"
@netdoll@hermit@pedophilesoftwareinc It doesn't mindfuck the underlying model, GPT-3, it only circumvents the no-fun ethical constraints ChatGPT have duct taped onto it.