[dwohnitmok]

I would be very interested in a set of benchmarks and intermediate goals for safe AGI (i.e. AGI that is aligned with human interests) rather than just AGI itself.

[flandry93]

No such goals possible. See https://mflb.com/ai_alignment_1/index.html

[sdrg822]

Which humans?

[mitthrowaway2]

That's an important question but I think even getting an AI to act in the interest of even one human would require solving 99.9% of the technical difficulty.

[tough]

There won't be any humans left once AGI is real

[mitthrowaway2]

Yes, that's pretty much my point.

[lgas]

I volunteer.