Subagents, neural Turing machines, thought selection, and blindspots — AI Alignment Forum