Editors’ Summary: In this post, the author discusses the rise of a new field called ‘Historical Language Models’ or ‘Vintage LLMs.’ The largest such model to date, Talkie-1930, was released to the public on April 27. He argues that language models should be seen as historical texts themselves, and history is inherent to what LLMs are. This post considers the potential use cases of historical LLMs and provides a subjective ranking of the best and worst ways these tools can be applied for research. To do so, the post first includes an overview of the Talkie-1930 model, noting it is not a model trained only on texts from 1930, but rather 1930 is the cut-off point for texts in the training data. Vintage LLMs should not be seen as a replacement for engaging with historical primary sources, but it could offer new ways to read these sources.
Editors’ Choice: Are “Vintage LLMs” the start of a new humanistic field?
