Conversation
|
|
||
| When engaged in discussion (code review, bugs, etc), a human reply must get a human reply. Even in cases where an AI agent initiates a conversation (e.g. by filing a bug or writing a PR description), its human operator is responsible for the content of its output, including its correctness and quality. | ||
|
|
||
| This policy borrows heavily from the policies of [LLVM](https://llvm.org/docs/AIToolPolicy.html) and [Chromium](https://chromium.googlesource.com/chromium/src/+/main/agents/ai_policy.md). |
There was a problem hiding this comment.
Can you wrap all these paragraphs at 80 cols.
|
|
||
| Contributors may use whatever tools they would like to craft their contributions, but there must be a human in the loop. Contributors must self-review and understand all code and documentation updates (with or without AI tooling) before sending them for review to ensure the correctness, design, and style meet the standards of the project. The contributor is always the author and is fully accountable for their contributions. Contributors should be sufficiently confident that the contribution is high enough quality that asking for a review is a good use of scarce maintainer time, and they should be able to answer questions about their work during review. To aid reviewers, contributors should flag areas they are not confident about that had AI assistance. This can be done in PR comments, the PR description, or in code comments. | ||
|
|
||
| Contributors must attest that the code they submit is their original creation, regardless of whether AI tooling was used. |
There was a problem hiding this comment.
Can one do this? If the AI wrote your code is it really your original creation? Is this language that is used in the Chromium or LLVM policy?
There was a problem hiding this comment.
I assume this statement from the Chromium policy is meant to ensure that copyright protections still apply, but IANAL.
|
|
||
| When engaged in discussion (code review, bugs, etc), a human reply must get a human reply. Even in cases where an AI agent initiates a conversation (e.g. by filing a bug or writing a PR description), its human operator is responsible for the content of its output, including its correctness and quality. | ||
|
|
||
| This policy borrows heavily from the policies of [LLVM](https://llvm.org/docs/AIToolPolicy.html) and [Chromium](https://chromium.googlesource.com/chromium/src/+/main/agents/ai_policy.md). |
There was a problem hiding this comment.
How does it differ from those mentioned policies? Can we perhaps just link to one of them?
There was a problem hiding this comment.
If we want AI agents to read this (i.e. to refuse to send a PR without a human reading it), it seems less likely that they would follow a link?
On that note can we can test what happens if an agent is instructed to write a PR without any human feedback with this CONTRIBUTING.md?
There was a problem hiding this comment.
I basically smashed together sentences I liked or that seemed important from both policies, dropped redundant or irrelevant parts, and slightly updated some wording. I agree that having a short policy in the repo rather than just a link is likely to be helpful.
There was a problem hiding this comment.
On that note can we can test what happens if an agent is instructed to write a PR without any human feedback with this CONTRIBUTING.md?
That would be interesting. I guess we would want to try both before and after landing this.
This branch has not been deployed
4 participants
No description provided.