Trained LLMs Exclusively on Pre-1913 Texts

Researchers have trained large language models (LLMs) using only pre-1913 texts, aiming to explore the capabilities of historical language models.

Details

Researchers have developed a set of large language models (LLMs) trained exclusively on pre-1913 texts, aiming to explore the capabilities of historical language models. By training on a corpus of texts from before the 20th century, the researchers sought to understand how well these models can capture and generate language from a bygone era. The historical LLMs were then compared to modern LLMs trained on contemporary data to assess their relative performance and capabilities. This research provides insights into the evolution of language and the potential applications of historical language models in areas such as digital humanities, historical analysis, and language preservation.