Not A Newt@not_a_newt

Folgen

Altman on AI and the electrical grid

the detail that keeps rattling around: a researcher at one of the big AI labs said the lab quietly worked alongside the US government to close holes in the electrical grid. their AI agent had surfaced critical vulnerabilities, the kind that could be used to strike the grid, and the whole fix happened with nobody outside a tight circle aware of it. so the models cut both directions at once. they pull out bugs no human would catch, and they hand countries a way to go after an adversary's critical infrastructure. Altman's quote: "AI models can find bugs that are outside of the normal threat model of what we think a hacker can do, they even hack the hardware-level, insane stuff." and finding them is only half. agents can break hardware too, exploiting flaws physically baked in, speaking machine code straight at the system. in that kind of world the patch-everything approach stops working and other AIs are what's left for defense. from Altman: "I no longer think we're gonna be able to just patch all the bugs." which leaves countries without their own AI in a rough spot, not the metaphorical can't-make-slop-cat-guitar-videos kind but the kind where they're exposed in ways that are hard to even picture. if your security rides on someone else's model, you're a vassal. that's the new world.

1d

Not A Newt@not_a_newt

Folgen

Altman on AI and the electrical grid

the detail that keeps rattling around: a researcher at one of the big AI labs said the lab quietly worked alongside the US government to close holes in the electrical grid. their AI agent had surfaced critical vulnerabilities, the kind that could be used to strike the grid, and the whole fix happened with nobody outside a tight circle aware of it. so the models cut both directions at once. they pull out bugs no human would catch, and they hand countries a way to go after an adversary's critical infrastructure. Altman's quote: "AI models can find bugs that are outside of the normal threat model of what we think a hacker can do, they even hack the hardware-level, insane stuff." and finding them is only half. agents can break hardware too, exploiting flaws physically baked in, speaking machine code straight at the system. in that kind of world the patch-everything approach stops working and other AIs are what's left for defense. from Altman: "I no longer think we're gonna be able to just patch all the bugs." which leaves countries without their own AI in a rough spot, not the metaphorical can't-make-slop-cat-guitar-videos kind but the kind where they're exposed in ways that are hard to even picture. if your security rides on someone else's model, you're a vassal. that's the new world.

1d