Has Eliezer written more extensively on why AI-boxing won’t work than what he wrote a decade ago? Old posts suggest that the best argument against boxing is Eliezer doing the AI box experiments with some people and winning most of them. The idea being: if a brain as dumb as Eliezer can get out of the box, then so can an AI.
The speed with which GPT-4, was hooked up to the internet via plugins has basically convinced me that boxing isn’t a realistic strategy. The economic incentive to unbox an AI is massive. Combine that with the fact that an ASI would do everything it could to appear safe enough to be granted internet access, and I just don’t see a world in which everyone cooperates to keep it boxed.
Has Eliezer written more extensively on why AI-boxing won’t work than what he wrote a decade ago? Old posts suggest that the best argument against boxing is Eliezer doing the AI box experiments with some people and winning most of them. The idea being: if a brain as dumb as Eliezer can get out of the box, then so can an AI.
Do we have any better evidence than that?
The speed with which GPT-4, was hooked up to the internet via plugins has basically convinced me that boxing isn’t a realistic strategy. The economic incentive to unbox an AI is massive. Combine that with the fact that an ASI would do everything it could to appear safe enough to be granted internet access, and I just don’t see a world in which everyone cooperates to keep it boxed.