The Natural Abstraction Hypothesis: Implications and Evidence — LessWrong