With almost 70 million monthly ChatGPT users in the U.S., substantial language models and chatbots are increasingly utilized for daily tasks, corporate initiatives, research techniques, and beyond. Nevertheless, despite this swift growth, the methods behind the training of large language models and the reasons for their specific responses to prompts remain ambiguous.

A duo of researchers from Florida State University, comprising a computational linguist and a philosopher of science, is embarking on a groundbreaking research field to explore why ChatGPT generates its responses. Their investigation, “Why Does ChatGPT ‘Delve’ So Much? Exploring the Source of Overrepresentation in Large Language Models,” was released in the Proceedings of the 31st International Conference on Computational Linguistics this January and serves as the inaugural scientific evidence that highlights why ChatGPT tends to excessively utilize particular words.

“We were motivated to undertake this research after noting conversations online among academics who observed various abstracts overly relying on certain specific words,” stated Zina Ward, assistant professor in the Department of Philosophy and instructor in the Interdisciplinary Data Science Master’s Degree Program. “Alongside identifying the words that were overly represented by ChatGPT, our investigation aimed to comprehend the underlying reasons for this excess. These language models are so new that there isn’t much existing research analyzing their linguistic behaviors.”

The article “Why Does ChatGPT ‘Delve’ So Much?”: FSU researchers begin to uncover why ChatGPT overuses certain words    was first published on Florida State University News.