Verbosity decreases accuracy in large language models

Amusingly, talk like caveman being discussed here: https://news.ycombinator.com/item?id=47647455

Where we end up because of this scares me. Literally going back to caveman speak

See? Already doing it.