研究人员利用前沿大语言模型在历史档案研究中取得了进展。1GPT-6 Astra、Opus 5.5等AI工具通过与历史学家的协作，在密码破译、追踪炼金术知识传播和解密历史文件等领域展现了能力。1
其中一项成果是，Opus 5.5从塞缪尔·哈特利布的5000多份档案文件中发现，牛顿和哈特利布使用了匹配的密码和字谜来表示关键炼金术成分——匈牙利硫酸。1同时，研究人员利用模型分析识别了牛顿翻译的法文炼金术文本来源，发现了新的历史联系和证据。1
在密码破译方面，GPT-6 Astra成功破译了1941年7月10日的谜机消息，关键突破在于发现了德国联邦档案馆中可用的完整信息范围。1Opus 5.5则部分破译了两封16世纪西班牙皇帝查理五世的秘密代码信件，尽管后来发现这些信件实际上已在1530年代和1916年被破译过。1
模型分析还表明爱德华·凯利（John Dee的抄写员）在特定日期后变得越来越懒惰，开始更频繁地重复自己的写作模式。1研究者建议AI实验室和历史基金会应系统资助历史学家、档案馆员与AI研究人员的协作，以进一步推进这一领域的研究。1
Researchers are leveraging advanced large language models to unlock new historical insights through cryptographic analysis and archival investigation. In a series of experiments, AI models including GPT-6 Astra and Opus 5.5 have demonstrated significant capability in decoding historical documents and uncovering previously unknown scholarly connections 1.
GPT-6 Astra successfully deciphered an Enigma message from July 10, 1941, with the breakthrough emerging from access to comprehensive records held in the German Federal Archives 1. In a parallel investigation of Samuel Hartlib's collection of over 5,000 archival documents, Opus 5.5 identified that Newton and Hartlib employed matching cipher systems to represent key alchemical substances, specifically Hungarian sulfuric acid 1. Analysis of Edward Kelly, the copyist of John Dee, revealed patterns in the documents suggesting increased repetition in his work after a certain point 1.
The research also addressed 16th century diplomatic correspondence, with Opus 5.5 partially decoding two secret-code letters from Holy Roman Emperor Charles V, though subsequent investigation found these had already been deciphered in the 1530s and again in 1916 1. Researchers propose that AI laboratories and historical foundations should systematically fund collaborative efforts between historians, archivists, and AI researchers to further such investigations 1.
评论
还没有评论，欢迎留下第一条。