Note on algorithms with multiple trained components — AI Alignment Forum