zlacker

> ..is necessary because an AI could hack it's way out of a laboratory. Yet, they cannot explain how an AI would accomplish this in practice.

I’m sympathetic to your position in general, but I can’t believe you wrote that with a straight face. “I don’t know how it would do it, therefore we should completely ignore the risk that it could be done.”

I’m no security expert, but I’ve been following the field incidentally and dabbling since writing login prompt simulators for the Prime terminals at college to harvest user account passwords. When I was a Unix admin I used to have fun figuring out how to hack my own systems. Security is unbelievably hard. An AI eventually jail braking is an eventual almost certainty we need to prepare for.

>>simonh+(OP)
It’s less about how it could be hacked and more about why an AI would do that or have the capability to do it without any warning.

>>elil17+uM1
That’s the alignment problem. We don’t know what the actual goals of an AI trained neural net are. We know what criteria we trained it against, but it turns out that’s not at all the same thing.

I highly recommend Rob Miles channel on YouTube. Here’s a good one, but they’re all fascinating. It turns out training an AI to have the actual goals we want it to have is fiendishly difficult.

https://youtu.be/hEUO6pjwFOo