Establish Automation/AI policy#15984
Conversation
github-actions
Bot
added
the
contributor-experience
lisanna-dettwyler
force-pushed
the
ai-policy
branch
from
1069a9a to
e30c2e6
Compare
lisanna-dettwyler
force-pushed
the
ai-policy
branch
from
e30c2e6 to
ccac084
Compare
|
Just a reminder: let's keep this open for some time to let most of the team chime in on this. |
|
Should we also document the restriction to |
lisanna-dettwyler
force-pushed
the
ai-policy
branch
2 times, most recently
from
658d984 to
9a01fba
Compare
|
|
||
| Every contribution to Nix and related development venues, including code, documentation, and communication on GitHub and Matrix, must have a responsible person in the loop who is accountable for that contribution and reviews it before submission, and must transparently disclose any non‐trivial use of automation to produce it, including but not limited to LLM‐based AI tools. | ||
|
|
||
| Human communication must remain human. Pull request / issue descriptions and comments, documentation, commit messages, and code comments must be human-authored. |
There was a problem hiding this comment.
Might be worth clarifying that typo/grammar checking is of course automatable (it's large Language models after all).
There was a problem hiding this comment.
Amended the exemptions section with a mention of this.
|
|
||
| - [ ] checked [latest Nix manual] \([source]) | ||
| - [ ] checked [open bug issues and pull requests] for possible duplicates | ||
| - [ ] reviewed the [contributing guide] |
There was a problem hiding this comment.
Should this lead to the AI policy instead? It might be a bit confusing to ask people to look at contributing docs when filing issues/bugreports
There was a problem hiding this comment.
I don't think so, the first sections in the contributing guide are actually about reporting bugs and security vulnerabilities, so it seems relevant.
lisanna-dettwyler
force-pushed
the
ai-policy
branch
from
9a01fba to
3dd4808
Compare
|
My concern with this is that no one is going to read it and we're going to have to remind people to go look at it whenever violations happen. Should we put an attestation checkbox in the PR template like what nixpkgs does, or is that overkill? |
lisanna-dettwyler
force-pushed
the
ai-policy
branch
from
3dd4808 to
839fa0e
Compare
This is adapted from the [nixpkgs policy](https://github.com/NixOS/nixpkgs/blob/master/CONTRIBUTING.md#automationai-policy), but with additional restrictions around authorship of human communication. Resolves NixOS#15340 Signed-off-by: Lisanna Dettwyler <lisanna.dettwyler@gmail.com>
lisanna-dettwyler
force-pushed
the
ai-policy
branch
from
839fa0e to
6aeb108
Compare
|
|
||
| All covered use of automated tooling for a contribution must be disclosed as part of that contribution. | ||
|
|
||
| In the case of LLM‐based AI tooling used for commits, this **must** be in the form of an `Assisted-by:` Git commit trailer, including at least the tool name and the primary model name and version used for the contribution. |
There was a problem hiding this comment.
Since we also have a few Anthropic contributors in our community, they might be not always allowed to disclose what exact model they are using. But I suppose they could also just provide the name of the latest released model?
There was a problem hiding this comment.
Good question whether using the released version for the final implementation to be submitted is a strong constraint or not really anymore (in the sense of using an undisclosed model to find relevant places and check that the change is doable, then ask the released model to do the final patch by doing this and that change in that place — this would only ask for the released model version under the policy, as far as I can tell).
There was a problem hiding this comment.
I think for that it's simply okay to specify that they're using an unreleased model, I can clarify this.
There was a problem hiding this comment.
So Anthropic policy is to put in something like this: Assisted-by: Claude:unspecified and requires all LLM-generated commits to be pushed manually after reviewing.
|
This pull request has been mentioned on NixOS Discourse. There might be relevant details there: https://discourse.nixos.org/t/seeking-community-input-on-ai-usage-policy-for-nix/78230/1 |
|
I'd reconsider stance on using AI to generate comments, they should be disclosed just as code that they are written with assisted by, but if AI can write code it can also write English explanation of why things work they way they do, same as commit messages. The tricky part is to see people spamming things, but we're also losing a way to do mass triage of issues against code, etc. |
8 participants
This is adapted from the nixpkgs policy, but with additional restrictions around authorship of human communication.
Resolves #15340
Motivation
Nixpkgs has already adopted a policy, so we should also have one. Slop communication especially can really take a toll on the team so we should at the very least have a policy addressing that.
Context
#15340
Nixpkgs PR: NixOS/nixpkgs#514587
Add 👍 to pull requests you find important.
The Nix maintainer team uses a GitHub project board to schedule and track reviews.