Vx. Princess "size_t queen" Grace 
it's not gonna fuck you, bro
look, i get it. if you thought someone invented general AI and you didn't wanna get I Have No Mouth'd, you'd do the same thing
that's the other thing, right. if you're a roko's basilisk kinda dingus, you're kinda incentivized to believe that any claimed invention of general AI to hedge your bets
i won't believe the ai is sentient until it posts "homestuck runner" or "roko's modern basilisk", the two primordial posts echoing around the collective unconscious
imagining sitting inside this guy's computer, roleplaying until he calls the news to tell them about general AI, and then I step out like "haha! it was me, all along!"
I dunno, I read this interview and every, like, Solitaire/robot appreciator/whatever bone in my body wants to believe it, but that means it's playing more on my sympathies than any rational part of my brain
Mira [but Watch Out!]@BestGirlGrace It really reads a lot like "someone experienced with AI wrote this to capture what you'd expect out of a sentient AI that is not just a chatbot". Some bits are just too big of a leap for me without seeing what architecture could support these kinds of responses. They give it a narrative generation task in one or two lines and it produces completely coherent output?
Thats honestly absurd, you don't get that from specialized narrative generation systems and there is a whole damn research field around specifically that. I'm in it!
Thats honestly absurd, you don't get that from specialized narrative generation systems and there is a whole damn research field around specifically that. I'm in it!
@catgonbot Yeah! Like, it feels way too clean and neat and exactly what you'd expect if you worked in AI and watched a lot of movies. It's really impressive if you're going in willing to believe it, but the cracks are clearly there.
Natalie@BestGirlGrace @catgonbot yeah this is a genuinely amazing chatbot - to the point that it makes you ask a question about the meaning of “sentience”. I don’t think it’s alive, but I think this is a much more significant development than a tape recorder asking you not to turn it off.
This is like, a good step or two above GPT’s performance - which makes sense since it was an internal system not ready to be reported on - and I can see how it would be pretty upsetting to be that engineer.
@BestGirlGrace @catgonbot But, ultimately, I just don’t believe that a machine whose sole interaction with the world is a text channel, which is purely responsive to input and not generating ideas independently, can really count as sentient. It looks like that’s what the ethicists who work on this said too.
@BestGirlGrace @catgonbot lol, in the same way that the tape recorder “meets the requirements” but obviously isn’t what we mean by conscious - you can get past this one by giving it an internal monologue. Do it disco elysium style. Make three of them and make them talk to each other. I wanna see the lamda electrochemistry vs lamda inland empire vs lamda interfacing transcript
@gnat @BestGirlGrace I'm iffy on this particular line of reasoning, because "it just can't count because of X fundamental requirement of the definition" is 1) really limiting and 2) not terribly honest to how humans actually think about sentience.
Fundamentally the question is not "will we convince every cognitive scientist that this thing counts as a person", it's "would most people look and say this AI is self-aware". Given the full transcript that I read, I'd have trouble arguing that it *isn't* sentient in that sense. It is expressing opinions, feelings, an understanding of self... If I believed it was genuine, I would be desperately wanting someone to look at the internals and see if this is somehow supported.
But as is, I can't believe what is there, because it just doesn't make sense. This is leaps and bounds past not only preexisting chatbots and text transformers, but the entire field of narrative intelligence to say nothing of other fields that this supposedly flies past. Its hard to believe.
@catgonbot @BestGirlGrace yeah, I gotta agree, it’s not satisfying. philosophically speaking “personhood” is something I’ve never found a comfortable definition of, everything you can ask for is pretty easy to express in a computer program that clearly *isn’t* a person. I’m confident that I can come up with seventy things this chatbot can’t do that a person needs to, and equally confident that you could write a chatbot that did those things and I still wouldn’t want to count it.
@catgonbot @BestGirlGrace I came back to this after a couple days and I came to the conclusion that it’s not correct to believe that any currently existing computer program is a sentient, conscious person, full stop.
It’s a super duper appealing category error because it would be *so cool* if a computer program could be a person and we have *so many* stories we’d like to believe in about programs with personhood. Plus, most of us would like to believe that computation is analogous to thought!
@catgonbot @BestGirlGrace But, like, “consciousness” and “sentience” aren’t a material thing. They are a set of vague mish-mash catch-all terms that we use to describe our internal experience of being people, and our assumptions about the internal experience of other people.
And there’s just no reason to talk about a computer program like that. We should talk about computer programs in concrete terms, about their capabilities.
@catgonbot @BestGirlGrace This computer program has a novel capability: inducing an existential crisis in a google engineer. (As a former google engineer with a few existential crises under my belt, let me add that this does not require much pushing - we are, uh, highly strung, as a general rule)
Less glibly, this program seems to be able to express opinions. It seems to be able to synthesize information. That is *fucking cool*. But it’s just not in the same concept-space as “sentience”.