ChatGPT vs the 2-4-6 Task — LessWrong