Synthesizing Standalone World-Models, Part 1: Abstraction Hierarchies — LessWrong