大型语言模型的紧急自主科研能力（英文版）

英文标题：Emergent autonomous scientific research capabilities of large language models中文摘要：本论文展示了一个结合多个大型语言模型的智能代理系统，可自主设计、规划和执行科学实验，并通过三个不同的实例展示代理的科学研究能力，最为复杂的是成功执行加催化交叉偶联反应。最后，讨论了这种系统的安全影响，并提出了防止滥用的措施。英文摘要：Transformer-based large language models are rapidly advancing in the field ofmachine learning research, with applications spanning natural language,biology, chemistry, and computer programming. Extreme scaling and reinforcementlearning from human feedback have significantly improved the quality