Are you in the process of creating tasks that behave as an acid test for AGI? If not, do you think such a task is feasible? I read somewhere in the ARC blog that they define AGI as when creating tasks that is hard for AI but easy for humans becomes virtually impossible.