Subagents, trauma and rationality — AI Alignment Forum