No way this is the final benchmark.
Given access to mouse, keyboard and screen data, can the ai:
hold down a paid online job for 6 months
buy a pound of drugs on the darkweb and get away with it
gank a noob in WoW
do taxes for anyone in any country
duplicate Facebook
open and run an online business
design a racecar
do everything that a human did on a computer ever
By this ruler most humans aren’t GIs either. And if it passes the bar, then humans are indeed screwed and it is too late for alignment.
No way this is the final benchmark.
Given access to mouse, keyboard and screen data, can the ai:
hold down a paid online job for 6 months
buy a pound of drugs on the darkweb and get away with it
gank a noob in WoW
do taxes for anyone in any country
duplicate Facebook
open and run an online business
design a racecar
do everything that a human did on a computer ever
By this ruler most humans aren’t GIs either. And if it passes the bar, then humans are indeed screwed and it is too late for alignment.