[docs] add a policy for AI assisted contributions

[odow]

Please leave comments.

https://jump.dev/JuMP.jl/previews/PR4138/developers/ai_policy/

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.91%. Comparing base (9c52bb9) to head (55a879b).
⚠️ Report is 5 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #4138   +/-   ##
=======================================
  Coverage   99.91%   99.91%           
=======================================
  Files          42       42           
  Lines        6224     6224           
=======================================
  Hits         6219     6219           
  Misses          5        5           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[mlubin]

The scipy/sympy policies already look pretty good, maybe we should use those directly as the starting point? The text is tight and I think we could adopt it with only minor edits.

[odow]

My opinion on the SymPy one: https://docs.sympy.org/dev/contributing/ai-generated-code-policy.html

• I don't like the disclosure rule. There's no way to enforce this, and it's just going to become everyone. We don't require disclosure if you used stack overflow, for example.
• Code quality: this is subjective. We don't want bad PRs, but I think equally, we won't want very large PRs even if they are high quality.
• Copyright is also impossible to enforce or judge. How am I meant to know if they violate someone's copyright?
• Communication is the thing that is most important to me.

[jd-foster]

I would prefer "GenAI" (or even Generative Algorithms, although obscure) to just "AI" in this context of code contributions.

[odow]

I've added the same disclaimer as SciPy: https://scipy.github.io/devdocs/dev/conduct/ai_policy.html

Another point to discuss: the responsibility, code quality, and copyright issues: these apply to all contributors all the time. It seems weird to call them out in an AI policy. Copyright considerations apply to anyone submitting code at any time. And the code quality: SciPy says "we will reject pull requests that we deem being AI slop" but why wouldn't they reject any pull request that is low quality?

I wonder if we need a separate policy and not just:

1. write in your own words
2. small changes are better than big changes
3. open an issue to discuss before opening a PR (see 1)

[mlubin]

• The disclosure rule for code is becoming pretty standard both in the open source world and within companies, so I don't think it's a big ask.
• Agreed, we should broaden the code quality point into considerations on maintainability of the contribution: By contributing a new feature, you're implicitly asking the core devs to maintain the feature. Consider an explicit discussion in the PR about the maintenance plans for the code if applicable. This isn't specific to AI but is aggravated by vibe coding.
• The explicit prohibition on agent-submitted PRs sounds like a good idea.
• The copyright discussion is important. I don't think it hurts to remind people using GenAI about it.

[odow]

Many changes to bring it much closer in line to SciPy.

[odow]

How about now

[odow]

Here's a case:

jump-dev/DiffOpt.jl#356 (comment)

[blegat]

Totally agree