Letters are welcome via email to letters@economist.com Find out more about how we process your letter
欢迎通过电子邮件投稿至 letters@economist.com。了解更多关于我们如何处理来信的信息。
I work for an AI lab as a scientist and read your piece on how AI companies are not merely providing a tool for discovery but choosing to wield(运用；行使（权力、工具）) it themselves (“A scientific revolution”, October 17th). My field of palaeontology uses large language models to investigate the evolution of fossils by mining the formal descriptions of species from literature that often extends back to the 18th century.(我的古生物学领域利用大语言模型，通过挖掘可追溯至18世纪的文献中对物种的正式描述，来研究化石的演化。) How fossils are formally described (their taxonomy(分类学；分类体系)) is key to their identification and use.
我在一家AI实验室担任科学家，读到了你们关于AI公司不仅提供发现工具、还选择亲自使用该工具的文章（《科学革命》，10月17日）。我的古生物学领域利用大语言模型，通过挖掘可追溯至18世纪的文献中对物种的正式描述，来研究化石的演化。化石的正式描述方式（即其分类学）是识别和使用化石的关键。
The trouble is that LLMs are not good at understanding very old literature and how terminology(术语；专业词汇), measurement and concepts of species and evolution have changed over two centuries. This means an enormous amount of work for human scientists in cleaning the data before the LLM can be put to work(投入使用；开始工作). It’s the same for other descriptive, observational sciences like botany and zoology, where written descriptions and evidence are a big part of the accumulated knowledge of humanity codified in legacy scientific literature.(植物学和动物学等其他描述性、观察性学科也是如此——在这些学科中，书面描述和证据是人类积累知识的重要部分，被编纂在传承下来的科学文献中。)
问题在于，大语言模型不擅长理解非常古老的文献，也不擅长把握术语、测量方法以及物种和演化概念在两个世纪里发生了怎样的变化。这意味着人类科学家需要做大量工作来清洗数据，之后才能让大语言模型投入工作。植物学和动物学等其他描述性、观察性学科也是如此——在这些学科中，书面描述和证据是人类积累知识的重要部分，被编纂在传承下来的科学文献中。
The Kane mutiny
凯恩哗变
Be prepared to help
做好准备去互助
Bells and Beefeaters
钟声与御林侍卫
Pick a number, any number
随便挑个数