x
Safely controlling the AGI agent reward function — AI Alignment Forum