Breaking the Optimizer’s Curse, and Consequences for Existential Risks and Value Learning — AI Alignment Forum