"the LLM group produced statistically homogeneous essays within each topic, showing significantly less deviation compared to the other groups" [LLM group produced statistically homogeneous essays]
The quantitative finding of reduced deviation in LLM essays directly supports the structural claim that AI tools narrow the space of possible outputs, reducing individual variation in cognitive expression.
Reasoning from this article
The NLP analysis reveals that LLM users converge on similar n-grams, named entities, and topic ontologies, while brain-only writers diverge significantly. This suggests that LLMs function as attractor states in cognitive-linguistic space, pulling users toward modal outputs. The homogenization effect extends beyond individual essays to group-level patterns, indicating a structural constraint on cognitive diversity when AI mediates knowledge work.