We're getting a bunch of AI generated PRs and issues - and while it's not inherently bad per-se, it is a waste of maintainer time if it's straight copy/paste or it's super clear that the author spent no time at all to review things themselves before sending the PR to us, and expecting us to find all the issues. Basically we need something we need to point to when we have to close a PR/issue or even temporarily or permanently ban a user who is just firing slop at us.
So we need an AI contribution policy as part of our contribution guidelines.
I do like the K8s one - https://www.kubernetes.dev/docs/guide/pull-requests/#ai-guidance
I am also inclined to add something to the effect of:
If the maintainers come to the conclusion that your engagement is primarily low-quality LLM based across issues, comments and pull requests, with no review or adjustment on the authors part, the following steps can be taken:
- A warning in the issue and/or pull request.
- If the warning is not heeded the PR or issue may be closed.
- In repeat or particularly egregious offenses, the author may be temporarily or permanently banned from the agones-dev organisation
Although I'd want to communicate that I don't want to stop people doing something like a single issue that's LLM generated and then adjusted via LLM -- that's fine (as long as you review and agree with it -- I've definitely done this). There's a time and place for the tool, and a time and place for the human to make sure there's oversight for the tool.
Thoughts?
We're getting a bunch of AI generated PRs and issues - and while it's not inherently bad per-se, it is a waste of maintainer time if it's straight copy/paste or it's super clear that the author spent no time at all to review things themselves before sending the PR to us, and expecting us to find all the issues. Basically we need something we need to point to when we have to close a PR/issue or even temporarily or permanently ban a user who is just firing slop at us.
So we need an AI contribution policy as part of our contribution guidelines.
I do like the K8s one - https://www.kubernetes.dev/docs/guide/pull-requests/#ai-guidance
I am also inclined to add something to the effect of:
If the maintainers come to the conclusion that your engagement is primarily low-quality LLM based across issues, comments and pull requests, with no review or adjustment on the authors part, the following steps can be taken:
Although I'd want to communicate that I don't want to stop people doing something like a single issue that's LLM generated and then adjusted via LLM -- that's fine (as long as you review and agree with it -- I've definitely done this). There's a time and place for the tool, and a time and place for the human to make sure there's oversight for the tool.
Thoughts?