Numberwang: LLMs Doing Autonomous Research, and a Call for Input — LessWrong