I’ve used it as a much smarter SO and Google for understanding existing code and getting suggestion how to solve some low-level problems in code as I’ve mainly working on fixing a lot of bugs (not my own code) and it’s in a programming language and libraries that I’ve got almost no previous experience in. It’s not always accurate but the amazing thing is that it’s somewhat steerable, but you need to have enough experience and understanding to detect subtle errors.
I’m only feeding it small snippets of code and sometimes we only chat about toy examples that are related to what I work on so I don’t have any major concerns about data leaks or hacks.
I’ve also tested it’s ability to do BDD, TDD, CI/CD and some more esoteric things such as formal verification with TLA+. My experience have been it works pretty good for any thing which have good amount of examples and related content. It’s actually a very good tool for learning as you can query it for issues you’re having while learning.
The only issue is it’s not up to date for leading edge stuff because of it’s cutoff time.