I find myself using QA on the codebase (not a generic LLM) more often. As a maintainer of internal and external facing python libs, I am wondering how much time I should spend maintaining working examples and docs.
I use LLMs as a last resort, when I can't find something in the official documentation first, unofficial documentation second, and the code itself third. And I assume that at least part of what a language model encodes about a library would've been derived from the documentation, and scanned code directly derived from that documentation, anyhow.
I usually start by looking at the docs for a non-trivial example or a Getting Started doc.<p>If those are insufficient, then I try to query an LLM, but that doesn’t always work depending on how recent the library is or how much the API of the library changes over time.