[Meta] Add AI policy to CONTRIBUTING.md#10345
Conversation
christinerogers
left a comment
christinerogers
left a comment
There was a problem hiding this comment.
This looks like a great comprehensive first policy change.
As always we might find things to tweak later but would be great to merge this asap.
This adds a policy on AI to the CONTRIBUTING.md file. The section on licensing is based on curl's policy and puts the burden on the contributor to ensure that ensure that generated code can be submitted. The rest is written by me as a first pass. I would personally prefer a stricter policy, but I think this is the minimum we need to be able to triage pull requests fairly and effectively.
|
I think the language around Discussions is too harsh and also somewhat confusing. We should alter that. And I think we should also add a Governance section with something like the following "Contributors must not input confidential LORIS data, credentials, PHI, or unpublished datasets into third-party AI systems." Otherwise, we can ratify this with time, but I think it's alright otherwise. |
christinerogers
left a comment
christinerogers
left a comment
There was a problem hiding this comment.
Suggesting (not requesting) these changes to move us forward from @samirdas 's feedback
- @driusan over to you to review / accept / adapt / raise for followup discussion e.g.Tuesday.
CONTRIBUTING.md
Outdated
| Usage of any large language model to reply to any user's issue, GitHub discussion, | ||
| or pull request comment is considered abusive and strictly forbidden. |
There was a problem hiding this comment.
Any use of generative AI tools used for communications is very strongly discouraged and considered deleterious to collaborative interaction with the team. Examples include but are note limited to: text in (comments on) GitHub issues or pull requests, responses to requests, or GitHub Discussion forum contributions. As such, usage of any large language model or otherwise AI-assisted generative text to reply or contribute on GitHub must be explictly mentioned alongside the text, to help the team in triaging.
There was a problem hiding this comment.
this suggested change attempts to :
- adapt per @samirdas' comments
- retain the spirit of @driusan's original text
There was a problem hiding this comment.
I don't think this retains the spirit. The problem with LLM replies to comments isn't triaging and "flagging" it doesn't help. We've already had people respond to reviews with copy/pasted replies from (presumably) ChatGPT. I'm not a fan of the word "deletrious" either (I don't know why, personal preference I guess.. just sounds weird to me).
How about:
Any use of large language models or generative AI tools in areas intended
for human communications or feedback is highly disruptive to collaborative
interaction with the community and forbidden. Examples include
but are not limited to: comments on GitHub issues, replies to feedback on pull requests, or GitHub Discussion forum contributions.
There was a problem hiding this comment.
Thanks dave, both your suggestions work for me.
@samirdas can you read what Dave just suggested in both of these comments above and let us know if that works for you?
Merging so this doesn't delay the release will come up pretty soon - Monday I think.
There was a problem hiding this comment.
How about this phrasing: "Any use of large language models or generative AI tools in areas intended for human communications or feedback can be highly disruptive to collaborative
interaction with the community and is strongly discouraged. Examples include
but are not limited to: comments on GitHub issues, replies to feedback on pull requests, or GitHub Discussion forum contributions.
There was a problem hiding this comment.
Works for me
@driusan any other fine tuning we should discuss in the Loris meeting, or do you have clear enough feedback to move forward for merging ?
@driusan if this works for you from what we agreed let's get this merged Co-authored-by: christinerogers <christinerogers@users.noreply.github.com>
christinerogers
added
the
Critical to release
5 participants
This adds a policy on AI to the CONTRIBUTING.md file.
The section on licensing is based on curl's policy and puts the burden on the contributor to ensure that ensure that generated code can be submitted. The rest is written by me as a first pass.
I would personally prefer a stricter policy, but I think this is the minimum we need to be able to triage pull requests fairly and effectively.