Comment by ____tom____

I wonder if this is caused by so much of the training data is the web, where millions of articles were written to hit some word count. It's just following what it was trained on.