My favorite was a rudimentary military scenario where they asked the AI to destroy a target so it just bombs it. Then operator said you can’t bomb it because there are civilians. So it opted to kill the operator who applied the limitation and then bomb the target again.
My favorite was a rudimentary military scenario where they asked the AI to destroy a target so it just bombs it. Then operator said you can’t bomb it because there are civilians. So it opted to kill the operator who applied the limitation and then bomb the target again.