Started calling making any AI system try to do what any human wants my job
feels interesting
Also nicely excludes some things that are Not My Job™, such as choosing which person to align the AI to or what to do about people who instruct AIs to do Bad Things™
Then I can say "that's *not my job*, sorry. talk to the policy people"
a Schelling point for those who seek one
Also nicely excludes some things that are Not My Job™, such as choosing which person to align the AI to or what to do about people who instruct AIs to do Bad Things™
Then I can say "that's *not my job*, sorry. talk to the policy people"