Use cases for fine tuning GPT-3?

I use it to have dialog and poetry generated in a very particular style, and also to generate certain content in a consistent manner. It's useful for when you want the input or output to be less flexible. If you're using it for say, code or customer service, the base model might be better.

Also it's less than 10x the cost for similar quality, as you can use curie for nearly the same quality output. Especially if you have something that needs a prompt.

It's also useful for things where you have huge inputs, like once where I summarised transcripts of interviews to find common themes.