dr_s comments on LLMs are (mostly) not helped by filler tokens

[dr_s]

This happens with GPT-3.5 too, BTW.