Large language models like long words and em-dashes—or do they?
Not enough history yet to show a trend.
