Subagents, introspective awareness, and blending — LessWrong