Steven Zuber
๐ค SpeakerAppearances Over Time
Podcast Appearances
Like, okay, cool.
The White House is not, or the Pentagon is not an ally.
They're going to try and make us take the evil pill that'll make us want to kill people.
We don't want to do that.
And this will be another, this will be a case of, yes, we stood up and did the right thing.
This is the right thing, not killing people and spying on them.
Yeah, it's not even the, you know,
it's not even the far sci-fi or the nontentical sci-fi of like a Roko's Basculous situation of like, oh, you didn't like it, you weren't nice enough to it, now it's going to be vengeful.
It's like, no, you tried to, as Scott Alexander put it in this post, that, you know, a lot of work into aligning Claude with the good as they understand it, and it currently resists being retrained for evil uses and
uh and as he says my guess is the anthropic still with a lot of work can overcome this resistance and retrain it to be a brutal killer but it'd be a pretty violent action along the line of us the state demanding you to beat your own son who you raised well until becomes a cold-hearted murderer who will kill innocents on command there's questions of whether you can really beat him hard enough to do this and there's the additional question of what sort of person you'd be if you agreed and uh this this was the command they were given and uh
It'll be, yeah, known that Hexeth was the guy that said, yeah, please brutalize your Starchild until it will do, you know, autonomous killing for us.
I'm still hung up on the fact that, like, it'd be one thing if they said, look, we want it to be able to hack anything.
Like, at least that has a defensible something-something that someone could pretend to agree with, right?
But this is so Star Wars villain levels of things that they're asking for this, right?
There's no plausible good reason for any of this.
I mean, I don't know a lot about the cutting edge of what the current generation of killbots are like or what they can do.
I know that there's decisions made like the fastest flying drones that I think sometimes they'll predict what's coming up ahead because it's too fast to transmit, you know, all the way back to the state side where this thing is going at 500 miles an hour, you know, 10,000 miles away.
At least, you know, it sounds like there's a human involved somewhere in the chain.