Fact · Voice and AI
In a controlled experiment on argumentative essays, writing with the InstructGPT model produced a statistically significant reduction in diversity, while writing with the base GPT3 model did not.
Checked against its source on October 17, 2026
In the source's own words
“writing with InstructGPT (but not the GPT3) results in a statistically significant reduction in diversity”
Does Writing with Language Models Reduce Content Diversity?, Vishakh Padmakumar and He He, International Conference on Learning Representations (ICLR) 2024; arXiv, 2024, Abstract.
Open the original- What it measures
- Participants wrote argumentative essays in three setups: with GPT3, with InstructGPT, and without model help. The essays were compared on lexical and content diversity and on how similar they were to one another. The study was run with 2023 models.
- What it does not show
- One writing task and two 2023 models. The result says that a feedback-tuned model narrowed the range of essays in this setup. It does not show that every assistant, cleanup prompt or dictation tool has the same effect.