Vector semantics and “Kubla Khan,” Part 2 — LessWrong