Probably yes - and I think my results suggest that optimizing for each would not only mean different endpoints but also different embedding models behind each. One model may measure best for short human-authored queries, while a different model may provide better results for longer LLM-authored query forms.