do have some kinds of plans that route through “build a [non-massively-superintelligent] thing they’re pretty sure they can figure out how to build and use it to help humanity
I think they have plans that look like “build the non massively superintelligent thing and hope this bails us out somehow mostly by having it do safety research fast enough (and trying to sure it has good enough judgment and alignment for this to be fine)”. I think to a large degree they think that you (very likely?) don’t need huge amounts of progress to align the AIs you get within a year or two of full automation of AI R&D (I disagree).
Idk how much this counts as a plan. I don’t expect they done a bunch of thinking like “Ah, well XYZ level of capability would suffice for this level of physicial tech progress, and that would suffice for slowing things down enough to be fine, and we can probably handle XYZ level of capability”.
I think they have plans that look like “build the non massively superintelligent thing and hope this bails us out somehow mostly by having it do safety research fast enough (and trying to sure it has good enough judgment and alignment for this to be fine)”. I think to a large degree they think that you (very likely?) don’t need huge amounts of progress to align the AIs you get within a year or two of full automation of AI R&D (I disagree).
Idk how much this counts as a plan. I don’t expect they done a bunch of thinking like “Ah, well XYZ level of capability would suffice for this level of physicial tech progress, and that would suffice for slowing things down enough to be fine, and we can probably handle XYZ level of capability”.