Why do you find yourself arguing with large language models during the drafting process?

Large language models are cautious by nature and tend to water down original theses. I have to push the model to respect my original vision rather than smoothing it out, ensuring that the final draft remains as sharp and distinct as my thoughts.