Approving reinforces low-effort behaviors — AI Alignment Forum