Re-review AI policy to what can reasonably be enforced once slop fest is over, if ever

[bbhtt]

Title.

[pinobatch]

In particular, do we want to be rejecting programs written in Python on grounds that CPython merged LLM-generated commits starting sometime after version 3.14a?

[bbhtt]

This issue isn't for external people to comment on, it's a reminder for myself and internal discussion.

You will find the answer in first line of #613 (comment) and that won't be changing.

[barthalion]

As I mentioned in the past, I'm against a straight ban. We should have an AI disclosure policy, where submitters explain how they used LLM for development of their app. Using it for code reviews, architecture discussion, suggestions or debugging is fine; minimal code generation of boilerplate should be acceptable too. Generation of substantial part of the code base would be rejected.

Essentially what IzzyOnDroid does (https://codeberg.org/IzzyOnDroid/repodata/issues/276) – minor/"chat" usage is allowed, what constitutes a slop isn't. Requires some extra review ping-pong to have the submitter say specific files that were tainted by LLM usage.

I assume this is where we want to draw the line to avoid another slop flood.

[bbhtt]

fwiw I wanted to move on this after a while, to gather some details about how the current policy is doing etc. Maybe it causes an uptick in proprietary apps that's net negative maybe not etc.

The slop flood is still ongoing too.

We can for sure ask to disclose but the main problem there is people will try to hide it when they have the chance and we need someone to review the claims and the codebase and classify it as substantial/ non-substantial etc.

My thinking (nothing final) was-

• fully generated is fully banned
• asistance is banned if it leads to sloppy practices or the assistance looks unreviewed.
• anything on the submission PR is still fully banned.

we don't care if they are using it in their IDE / personal capacity / PR review / issue reports / security reports etc.

If there are signs of LLM assitance (disclosed or found) we ask some for maturity period to make sure things will be maintained after the first month.

For FOSS "maturity period" could be maintaining the app for at least 6 months and having multiple releases, having more than 2 contributors etc.

I'm not yet sure how to define this for proprietary apps.

[ramcq]

Drive-by thought - could we adopt https://github.com/ghostty-org/ghostty/blob/main/AI_POLICY.md - I think it captures a lot of our sentiment here - we need honesty in the tool use being disclosed, we need human judgement to be exercised - control/review/edit, and we need our time not to be wasted by nonsense.

[bbhtt]

It captures the spirit but I think we want to be stricter in certain areas than that as on first read it felt a bit open.

[barthalion]

Here's my stab a this: #641