We support the use of AI tools to help prepare issues, pull requests, reviews, or comments. We expect everyone interacting with this repo to follow the below policy whenever they use AI tools.
AI-generated content in comments, issues, or PRs must be clearly disclosed and
contained (e.g. using a code block or a quote block). AI-generated content must
be accompanied by human commentary explaining its relevance. For example:
"Codex produced the following analysis: <insert codex output here>, so I
believe that <insert human analysis here>". The only exceptions to this rule
are the pytorchbot automations.
Please do not respond to comments+questions by pasting raw or lightly reviewed AI-generated text. Other users' comments+questions are requests for your understanding, reasoning, and judgment. Unreviewed AI output shifts the burden of verification onto the other party and can make discussions slower, noisier, or misleading.
For pull requests, please carefully read the code before submitting it for review, especially if AI tools helped write or rewrite it. Make sure the implementation is something you understand, that the important ideas are clear, and that the code is high quality. AI-generated code can sometimes be overly complex, indirect, inconsistent, or include artifacts that obscure the main idea. Before submitting, simplify where possible and make sure the core change is easy for maintainers to review. If your PR is not ready for review, please use the GitHub draft PR feature.
All contributions should involve a human who understands and can take responsibility for the work produced with AI assistance. We do not accept contributions created by fully autonomous agents, and we may close pull requests that appear to have been generated without meaningful human involvement.