Collection of GPT-3 results — LessWrong