Slowing AI: Foundations — LessWrong