Using LLMs to trace alchemical knowledge and decode 17th century letters (resobscura.substack.com)

Recent advancements in AI, marked by the releases of GPT-6 Sol and Opus 5.5, indicate a transformative potential for historical research. These models are now being applied not just for standard research assistant tasks like document transcription, but for tackling complex historical challenges. The author advocates for a collaborative approach where historians work alongside AI tools to unearth new insights, suggesting that previous hesitance among historical researchers to harness AI could change dramatically. This collaboration could lead to breakthroughs in interpreting historical texts, understanding coded communications, and discovering new connections between disparate scholarly works. The significance of this development lies in the models' ability to solve "tractable" historical problems — ones where expert identification has occurred, data is digitized, and solutions can be verified. For instance, GPT-6 Astra successfully decrypted a 1941 German message and identified previously unrecognized Latin translations from Isaac Newton’s references. The methodology employed by these models could enhance interdisciplinary research, allowing historians to investigate vast datasets across languages and time periods much more efficiently than human researchers alone. This evolution in AI capabilities could ultimately reshape the landscape of historical scholarship, opening new avenues for discovery and reinterpretation.