Gpt-3.5-turbo and frequency/presence penalty

[tom3]

Ah interesting! Some of the output would have been more than 400 tokens, but probably not by much - which could explain the subtlety. I will experiment with longer outputs.
Many thanks!