Curious how this performs against chatgpt with few-shot prompting. More and more evidence recently has shown that fine tunes/retrieval don't measurably improve

Curious how this performs against chatgpt with few-shot prompting. More and more evidence recently has shown that fine tunes/retrieval don't measurably improve performance over few-shot, but with this being such a specific domain/skill, I'd be surprised if it didn't far outperform a few-shot approach.