leogao's Shortform

leogao

This is a special post for quick takes by leogao. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.

i think some people underrate how convergently important the capacity to pause if needed is. even if through some miracle we align the AI that strictly dominates humans at alignment research, even that is not a sufficient win condition for making the future go well, because that AI could need a lot of time to solve alignment for its successor (or its successor could need a lot of time to align the successor’s successor, etc), and race dynamics forcibly limit the amount of time there is to do alignment even at that point.

This is why I personally think "compute verification" is the most important problem on earth right now: https://canaryinstitute.ai/blog/most-important-problem/

But "expanding the field, even at the cost of ideological purity" is a close second, IMHO

2drnickbone1mo

It’s certainly struck me that as AIs approach the level of AGI, they might have the same reasons to worry about ASI as we do, and may become (powerful) allies in helping us pause its development.

5leogao1mo

capabilities are spiky. they may potentially be much worse at politics than technical work

leogao's Shortform

8

Ω 4

leogao's Shortform

8

Ω 4