I feel a bit confused by these questions, but to take a stab at them anyway:
Are there projects (or even megaprojects) that we can undertake, that can increase probability of noticeable GDP impact before we run AI in dangerous regimes?
Sure. Encourage people to use the apply the best current AI techniques in actual applications. This leads to more applications before the dangerous regime. It also plausibly hastens the dangerous regime (because the dangerous thing gets deployed as an application more quickly).
Does getting noticeable GDP impact before we run AI in dangerous regimes, help to reduce actual likelihood of us running AI in dangerous regime?
One useful effect is that it increases the probability of a warning shot, and the severity of the warning shots that we do get. One harmful effect (of the action I suggested, not necessarily of the overall goal) is that it normalizes “deploy AI as fast as possible”, including risky / dangerous AI.
Yeah, I think you could do that, though I’d guess it isn’t as efficient as encouraging other people to do this (that has significantly more leverage, since you’re affecting many companies, not just one you make yourself).
I feel a bit confused by these questions, but to take a stab at them anyway:
Sure. Encourage people to use the apply the best current AI techniques in actual applications. This leads to more applications before the dangerous regime. It also plausibly hastens the dangerous regime (because the dangerous thing gets deployed as an application more quickly).
One useful effect is that it increases the probability of a warning shot, and the severity of the warning shots that we do get. One harmful effect (of the action I suggested, not necessarily of the overall goal) is that it normalizes “deploy AI as fast as possible”, including risky / dangerous AI.
I’d guess that the harms exceed the benefits.
Yeah, I think you could do that, though I’d guess it isn’t as efficient as encouraging other people to do this (that has significantly more leverage, since you’re affecting many companies, not just one you make yourself).