Write Like It's 1866: LLMs Relearn Telegraphese

Researchers at the University of California, Berkeley, and the Massachusetts Institute of Technology (MIT) have developed a new technique to teach large language models (LLMs) to mimic the style of 19th-century telegraph operators. The technique, called 'Telegraphese,' uses a dataset of historical telegraph messages to train the models. The goal is to enable LLMs to write in a style reminiscent of the past. The researchers used a dataset of over 150,000 telegraph messages from the 19th century, which they used to train the LLMs. The trained models were then tested on a set of prompts, and the results showed that they were able to produce text in a style similar to the original telegraph messages. The researchers plan to apply the technique to other historical styles, such as 18th-century literature.