Policy proposal: moratorium on LLM-generated issues/PR descriptions/etc.

[taobojlen]

I've noticed that at ROOST we have a lot of LLM writing as we collaborate on GitHub. Many (perhaps most?) of our issues, PR descriptions, and discussions are quite clearly written by Claude et al. -- and I am probably guilty of this myself on occasion!

I think this has a lot of negative effects, and I'd like to propose that we take steps to reduce this. One example could be banning outright LLM writing for issues, PR descriptions, PR comments, and GitHub discussions (while allowing copyediting and so on).

When I read an issue/PR/comment that's clearly from an LLM, it does a few things:

• I don't know whether the author has proofread it, or how closely, so I cannot trust the information given. I have to be extra diligent in verifying it and this takes time.
• I also don't know to what extent the author has actually engaged with the topic at hand, or whether they're repeating assumptions made by some LLM. As powerful as LLMs are, they still have many failure modes, and when it comes to understanding, I trust an engaged human far more than a machine.
• I don't get any sense of the author. I don't know who they are, how they think, what they believe, how they relate to the thing we're discussing, and so on. This kind of sucks for community building.
• LLMs tend to be overly verbose.
• If I wanted Claude's opinion I would ask Claude myself (and I often do!). It feels pretty disrespectful if I write a thought-out opinion on something, and someone pastes an LLM response.

The end result is that remote collaboration is a lot less enjoyable & productive than it could be.

There is some precedent for this policy:

Cloudflare

I just declared a moratorium against AI-written change descriptions (e.g. PR and commit messages, also issues/tickets) from my team.

I was writing change descriptions that were worse than useless to me as I tried to review PRs: outlining details of the code that could easily be seen by looking at the code, but omitting the higher-level framing needed to understand broadly what the code is doing.

Task

Keep conversations human - PR descriptions, review comments, and issue discussions should be written in your own words, not generated by AI. Code review is a discussion between people, not bots.

LLVM

To ensure sufficient self review and understanding of the work, it is strongly recommended that contributors write PR descriptions themselves (if needed, using tools for translation or copy-editing). The description should explain the motivation, implementation approach, expected impact, and any open questions or uncertainties to the same extent as a contribution made without tool assistance.

[taobojlen]

I'm going to fold this into https://github.com/orgs/roostorg/discussions/62 as further discussion there!