I don't relate, I'm currently designing a lower level data structure based on some papers and its been a pretty valuable rubber duck. I've only had one real ins

I don't relate, I'm currently designing a lower level data structure based on some papers and its been a pretty valuable rubber duck. I've only had one real instance of it being incorrect that wasted my time, higher ranked trait bounds in Rust can only be applied over lifetimes and instead it was suggesting I try to use them for non lifetime generics as well. That said I'm not asking ChatGPT to write code that I use directly, I'm basically asking clarifying questions about the papers and other implementations I have on hand, and that's similar to how I use it in most cases. For code generation I rely on co-pilot, which actually has the context of my codebase. Edit: I will say this because someone else made me think of it: I find this tech all super useful on my like 20k loc and smaller side projects which are all self contained where as when I tried co pilot at work on a fragmented big corp code base I found it comparatively lackluster.