A more systematic case for inner misalignment — AI Alignment Forum