I think a sincere belief in AI's danger is compatible with regulatory capture. They think AI is dangerous. Ergo, it should not be developed openly and widely, the Chinese way, but rather more like nuclear weapons, under close guard.
replyI think the idea has some merit, but it only buys time, as people will soon be able to train dangerously powerful AIs in secret. I think the government needs to force the labs to solve alignment before developing more powerful models. Then the question of how to counter a poorly-aligned, rogue model arises; it's guaranteed somebody's going to let one loose.
I doubt nuclear nonproliferation would've worked if U235 were 20x more abundant in nature, and didn't require much in the way of concentration via centrifuge. It would require a truly unprecedented level of international cooperation and surveillance to lock down AI development/deployment to a similar level as has been done with fissile materials.
replyBut this isn't an argument against their point. If nuclear were far easier to make then every government would be far more authoritarian to prevent terrorist style nukes all the time. The average person would fully allow it after a few local nukes went off.
replyAnd it doesnt really matter if it was actual terrorism or false flags, the end result for either is the same.
Sincere is an interesting adjective to describe a company that won't even admit to how many parameters their models are.
reply