Ask HN: We need a standard way to declare how much AI was used in a PR
There's a big difference between these situations:
- a PR that is quickly drafted by LLM based on a prompt
- a careful spec designed by humans, worked on by AI
- a feature built with a dev and LLM closely working together, with code reviews
- a feature with parts fixed up by AI, but 90% human written
- human written, checked with AI
etc
Is anyone moving towards a standard way of describing this? I'd love something that can go at the bottom of an communication / PR / proposal.
I believe that it isn't that important to quantify how much AI is involved in a PR. It's not possible to measure it other than asking and right now it isn't even a good metric. LLMs could be producing good code and a human could be producing really bad code. As somebody said, just ask if AI was in any way involved in the coding process and treat the code as such.
[ ] Was AI used in this PR in any way at all?Does using he free version of Gemini in http://www.google.com to find the correct page of the documentation count as using AI?
Sardonic but not useful.