Researchers at Stony Brook University and Salesforce AI Research are investigating the reasons behind professional writers' dissatisfaction with Large Language Models (LLMs). They propose a refined model to better align machine language with human expression.
LLMs have transformed artificial intelligence writing assistants, impacting scientific research, persuasive communication, and creative literary endeavors. Despite their potential, differences between LLM-generated and human-written text raise concerns about their ability to produce high-quality writing.
A study led by Assistant Professor Tuhin Chakrabarty from Stony Brook University's Department of Computer Science highlights these challenges. The research, involving professional writers, suggests ways to improve alignment between machine-generated content and human writing. The paper received nominations for Best Paper and Honorable Mention Awards at CHI 2025.
Tuhin stated, “One significant issue we noticed is that LLM-generated text often suffers from a lack of originality and variety.”
The prevalence of LLMs in writing tasks leads to "algorithmic monoculture," where content becomes homogenized. This results in repetitive expressions lacking rhetorical depth. Such texts often miss the complex narrative techniques inherent in human creativity, opting for "telling" instead of "showing."
The full story by Ankita Nagpal is available on the AI Innovation Institute website.