Holly ⏸️ Elmore
I actually think proponents of scalable oversight/superalignment approaches should be embarrassed. If they were serious about making it work, they’d be advocating for more time, but the real point of this is not to oppose AI dev, so they won’t.
Holly ⏸️ Elmore
Apparently this needs to be said a lot more loudly: Superalignment is not a plan to take us to a safe Singularity. It is a comforting story at this point, and tho I by no means think Pause victory will be easy, it is more realistic than this fable working out.