I had the opposite experience but I think it depends a lot on how you write the options. Vague labels gave me meh results. Once each option had an explicit "not for" list plus a couple of examples it got a lot better. That's for email sorting though, so pretty easy categories
"Not especially good at classifying" does sound like a serious drawback for a classifier
Welcome to AI hype.
do you have a benchmark name and results we can compare to other models?
I had the opposite experience but I think it depends a lot on how you write the options. Vague labels gave me meh results. Once each option had an explicit "not for" list plus a couple of examples it got a lot better. That's for email sorting though, so pretty easy categories