Add a page for general contribution guidelines, with a code of conduct and AI disclaimer

[Ivorforce]

[AThousandShips]

I'd add a note about:

• Proof reading anything you use it for (if you do use tools you should use them to help you write code, not to write code for you)
• If you use LLMs to help you describe things, make sure the output is actually readable and as brief as it can reasonably be, LLMs are ridiculously verbose and meandering, so descriptions written using it should be pruned

I'd also say that we should make it clear that LLMs should not be used to write code you couldn't write yourself, i.e. solve problems for you, and that regardless of the tool you use, you should never contribute code you don't understand yourself

I.e. that if you can't solve the problem without an LLM, you should probably let someone else do it, because it just adds noise and forces other people to check things for you instead of letting someone who knows what they're doing write a solution

[Ivorforce]

Good ideas, I've transcribed your suggestions and changed the format into a list of requirements.

[AThousandShips]

I think this covers it well, will try to do a style and form pass tomorrow

[Mickeon]

We may want to make a clearer distinction on what artificial intelligence this is referring to. AI is a huge field of computer science, but what we have the displeasure of encountering is, for the most part, the "machine learning", "generative", and/or "language-learning models" subset, so to speak.

[clayjohn]

Some other examples for inspiration (posted by Stuart on rocketchat):

• https://datafusion.apache.org/contributor-guide/index.html#ai-assisted-contributions
• https://github.com/ghostty-org/ghostty/blob/main/CONTRIBUTING.md#ai-assistance-notice

[clayjohn]

Suggested change

	- In particular, you must disclose if you used AI to write part of your contribution or issue / pull request description.
	- In particular, you must disclose if you used AI to write part of your contribution or issue / pull request description. Using AI translation to translate text that you originally wrote is always acceptable.

AI translation comes up 90% of the time when someone is accused of posting AI slop directly. There was some drama on Reddit a few weeks ago because someone with very little English knowledge used AI to translate and polish a nice write up they did. While I think it might go without saying that AI translation is fine, I think it is better to say it explicitly.

[Mickeon]

Indeed. We must be quite pedantic here and note that software like Google Translate are also part of the "artificial intelligence" umbrella term. It's not a language-learning model, but it uses roughly the same technology.

We must be clear here that using specialized software for the purpose of translating is fine, and... probably better than asking ChatGPT...

[Ivorforce]

I decided to rephrase your suggestion a bit. I wanted to avoid making it sound like we're welcoming people using AI to fill out our language gaps in weblate.

[clayjohn]

Suggested change

Contributions made entirely by AI are prohibited.

As a rule of thumb, only submit code that you fully understand and are prepared to explain to a maintainer. Maintainers are spending their personal time reviewing your code, it is your job to ensure that the code you submit is well-tested and functional. Please be respectful of their time and only submit code that you have put thought and effort into. Contributions made entirely by AI are prohibited.

I want to follow up the strict rules with a more general re-statement that clarifies why we don't want people to just dump fully AI authored code.

I am tempted to add something like "I don't want to bother reviewing code that you didn't bother writing" But I think that is a little to snarky for this sort of text.

[AThousandShips]

I think this is a good way to say that but professionally and politely

[Mickeon]

"Prohibited" itself is an excessively strong word. We don't often ban users for their low-quality submissions, nor we have the authority to prevent them from submitting them in the first place (which is what "prohibit" implies in my opinion).

[Ivorforce]

Generally I would agree, but in this case I think it may be appropriate. We don't want users setting up AI agents with the task of "Look through the Godot source code and submit fixes for bugs as PRs".

[AThousandShips]

I think it can be worth to make it clear that the expectations of understanding and being able to justify your choices for contributors and code applies to all contributions, regardless of use of AI, for example don't just copy code from some other source, even without attribution issues, without understanding it

AI is just an especially strong case of this as there's plenty of good reason to trust certain sources of code, but with AI there's plenty of cases of just garbage

[Ivorforce]

We may want to make a clearer distinction on what artificial intelligence this is referring to. AI is a huge field of computer science, but what we have the displeasure of encountering is, for the most part, the "machine learning", "generative", and/or "language-learning models" subset, so to speak.

That would be nice, but I'm not sure if I know of a way to do this. AI is pretty diverse, but I don't think there is any AI yet that we do want to be involved heavily in contributions. We don't want people to run AI agents to implement implementation proposals, we don't want people to use full-function autocomplete (without disclosing it anyway), and we don't want people to set up dependency bots or similar spamming the repo. So perhaps, in this case, using the buzzword "AI" does kind of work out?

I think it can be worth to make it clear that the expectations of understanding and being able to justify your choices for contributors and code applies to all contributions, regardless of use of AI, for example don't just copy code from some other source, even without attribution issues, without understanding it

AI is just an especially strong case of this as there's plenty of good reason to trust certain sources of code, but with AI there's plenty of cases of just garbage

I was planning to make a follow-up PR for a sub-section titled 'code from external sources' (or similar), where we would also explain the licensing thing.
We could re-iterate this point in that section as well.