Microsoft Research Paper Claims Sparks of Artificial Intelligence in GPT-4 — LessWrong