2010 (1)
2011 (1)
2013 (42)
2018 (128)
2020 (783)
2021 (1188)
2022 (1546)
2024 (1287)
其中一段摘錄：
AI Alignment and Safety
BG SHANE REEVES: You talk about alignment. Can you explain what you mean by alignment to help everyone understand?
ELON MUSK: Yeah. Just is AI going to do things that make civilization better, make people happy, or will it be contrary to humanity? Will it foster humanity or not? Will it be against humanity? So, obviously, we want an AI that will foster humanity.
I think in developing an AI to foster humanity… because I’ve thought about AI safety for a long time. Like, I’ve had probably a thousand hours of discussion about this. And my other conclusion is that?the best course for AI safety is to have an AI that is maximally truth-seeking, and also curious.?And if you have both of those things, I think it will foster… it will naturally foster humanity because if you want to see how humanity develops, humanity is more interesting than not humanity.
You know, I like Mars. I’m a big fan of Mars, obviously. And I think we should become a multi-planet civilization. I guess that’s very important. But the purpose of SpaceX is to make life multi-planetary. That’s the reason I created the company, and that’s the reason that we have the Starship development in South Texas. That rocket is far, far too big for just satellites. It’s an attempt to establish life on Mars, not just to send astronauts there briefly, but to build a city on Mars that’s ultimately self-sustaining.
But getting back to AI, if you’ve got a true-seeking AI that is naturally curious, my neural net, my biological neural net says that that’s going to be the safest outcome. Because if, like, if you wiped out Mars, this… you… Mars is not as interesting as Earth because there’s no human civilization there. Or thought of another way, if you want to render Mars, rendering Mars is pretty easy. It’s basically red rocks that look kind of like some parts of Arizona. You know? There’s not a lot of people. It’s just… it’s easy to render. It’s rendering, like, to Mars.
But rendering human civilization is much harder, much more complex, much more interesting. And so I think a curious truth-seeking AI would foster humanity and want to see where it goes.
全文：
