How LLMs Distort Our Written Language
Summary
The study analyzes how large language models distort written language across human-user, argumentative-essay, and peer-review datasets. It finds that LLM edits can shift conclusions, homogenize semantics, remove author voice, and alter stylistic and emotional content, with implications for science, culture, and everyday communication. The authors warn about the potential impact as AI writing becomes more integrated into professional contexts and call for caution and further research.