taylor

[taylor @taylor__ws]

@GoodExplain @Aella_Girl Exactly so! We want an AI that "cares" about us, but we just don't know how to do that. "Caring" is very fuzzy!

[taylor @taylor__ws]

@GoodExplain @Aella_Girl I agree with this! I fear the kind of AI that is beyond us in terms of ability-to-accomplish-goals *while yet* being amoral (or very morally impoverished).

[taylor @taylor__ws]

@GoodExplain @Aella_Girl It's more like: AI ends up with a weird, unintended set of goals/drives, pursues those aggressively, uses up lots of resources that we humans need, we can't compete, we peter out.

(It'll *understand* what "morality" is while it does all that, but why would it care?)

[taylor @taylor__ws]

@GoodExplain @Aella_Girl Idk maybe there are some people think AI will end up with a "morality bent on destruction" but that's very far from the mainstream model of why superintelligence destroys us!

[taylor @taylor__ws]

@darkzuckerberg_ @So8res 🤷‍♂️ just trying to help. I can tell you already have strong feelings about this stuff and I don't blame you, but "Nobody will answer this!" "Here's several answers!" "I'm not reading that propaganda, also you are lazy for not personally writing me an essay" felt kind of mean

[taylor @taylor__ws]

@darkzuckerberg_ @So8res It's weird to focus on whether I (random internet stranger you don't know) am personally full of shit, rather than on if the claims are true? All I've done is find for you what I think are useful sources if you're interested in the "how and why." Use them if you like.

[taylor @taylor__ws]

@darkzuckerberg_ @So8res Or if you prefer video:
AI-2027: https://t.co/uPMTCCehbr (or https://t.co/HB6bKVDcxd)
Yudkowsky/Soares: https://t.co/l8XJbqNbog
Interview with Holden Karnofsky: https://t.co/fgZVGq6gpz

(h/t @ohabryka for compiling)

[taylor @taylor__ws]

@MarcoAnton76652 @DKThomp The proposal is that US and China make a deal to both stop frontier AI development.

[taylor @taylor__ws]

@Fishwilsoncc @DKThomp The warnings are about the tech, and especially about the fact that we do not know how to have anyone be "in charge" of a superintelligence or to make it do what we want.

[taylor @taylor__ws]

@dmdebruijn I don't follow how the bubonic plague/nuclear war relate to how dangerous AI could be (could you clarify)?

A lot's been written on why we're very unlikely to be able to just "unplug" a superintelligence; main reason is it would try to stop you. https://t.co/p2zTC01ssr

[taylor @taylor__ws]

@jachiam0 @DavidSKrueger "as catastrophic"

Do you think it will be somewhat catastrophic (even if less catastrophic than David thinks)?

[taylor @taylor__ws]

@DalJeanis @ESYudkowsky Unfortunately, nobody knows how to do that; I'd bet something like that would be Eliezer's #1 choice too if we did know how

[taylor @taylor__ws]

@McaleerStephen That's right — we know that no one knows how to control a superintelligence, and that therefore we shouldn't build one. It is incredibly dangerous to race to ASI.

[taylor @taylor__ws]

@mcmoynihan I think sometimes people use "capitalism" to gesture at "any situation in which very rich people exist", not to describe the structure of the economy.

[taylor @taylor__ws]

@StopAntisemites Doxxing a random person because you disagree with their (presumed) beliefs is absolutely psychotic behavior.

[taylor @taylor__ws]

@LittleMammith This is the same as "Heads, God creates one person, tails he creates 1,000,001 people," right? I'm not sure what the jackets have to do with it.

[Isaac Luce @isaac_lewis]

@TylerAlterman "Supposedly a skilled musician can create a mind-blowing emotional experience by strumming a guitar. But we had 10 grad students strum guitars to 100 test subjects according to provided sheet music & saw no effect"

[taylor @taylor__ws]

@AdamBRozycki @Veeayvee1 @ESYudkowsky I'm curious why you're confident that AI that surpasses humans in "those aspects of human intelligence that enabled us to dominate the Earth" *won't* be developed? For instance, is there an example "aspect of intelligence" that you think AI won't acquire?

[taylor @taylor__ws]

@RichardSSutton Smart ≠ good.