What is LLM / AI policy for the Flow development process

[ghistrash]

Hi! / Привет!

Sorry for yet another such discussion on the WWW! Unfortunately, i am forced to think about it now.

I met Flow because decided to look for alternatives to Helix-Editor (Kakoune, Vim, Zed, etc..), whose position is uncertain still.

This is partly stupid, and escaping all the llm/ai oppression is unlikely – but this is about the path, not the goal. Maybe this is even a litmus test..

However, there are also less obvious positive consequences: a significant narrowing of the field of choice, paradoxically, is the key to new discoveries! And today i am knocking to a Flow's door...

[neurocyte]

Simply put, our AI policy is that there is no such thing as an AI policy.

People can own code (copyright), not AI. Any contribution you make to flow must be owned by you (the contributor) and be contributed under the MIT license that the project uses. If you want to use an LLM to generate code, that is up to you, but your contributions will be evaluated entirely as your own personal work and attributed only to you.

Several practical considerations for this (non)policy:

• Commit authorship should be attributed to you (a human), and you only, as this is copyright relevant. Do not include AI attribution in any commit as that may be viewed as an attempt to assign copyright ownership, and/or pass responsibility for any quality issues, to an AI company. This will get your PR insta-closed.
• You may mention machine generation techniques in your PR message (not in the commit) to allow us to gauge the human review effort we should invest in your work. But you don't have to.
• If at all possible, keep contributions small enough to trivially review. (AI assisted or otherwise)
• If you want serious human effort given to any non-trivial PRs (AI assisted or otherwise), state clearly your level of personal (human) investment in the change.
• Code review of human hand written code (and/or coaching) will be given absolute priority over reviews of anything machine generated.
• "This was generate by Claude, but it proves that {favorite_feature} is [possible|sensible|a great idea]" is not a justifiable reason for a PR merge. Please don't waste our time by making us read random LLM output. We would rather be working on flow.

An LLM is the digital equivalent of a chainsaw. If you can create fine art with a chainsaw, good for you! We welcome your contribution. But, if you attempt to swamp us with low quality AI slop that we could've just generated ourselves, you will get the very cold shoulder and we will likely just ignore or insta-close your "contributions".

[neurocyte]

I had no intention of spamming the repository with any kind of ai output, no!

I didn't think that was what you were implying, but, since you asked, I thought it would make sense to document my general opinion on AI submissions somewhere public.

Your position inspired me. A great reason to start learning Zig. I am glad, thank you!

Yes! Learn Zig! Zig is an extremely satisfying language to learn and the community is generally very welcoming.

[jobukkit]

Any contribution you make to flow must be owned by you (the contributor) and be contributed under the MIT license that the project uses.

Commit authorship should be attributed to you (a human), and you only, as this is copyright relevant.

Code inserted by an LLM is not written by "you", it was written by the people whose code the LLM was trained on, and thus it cannot be owned/copyrighted/MIT-licensed by "you". If this kind of code is allowed at all, then the project violates copyright and its license is meaningless.

[neurocyte]

Code inserted by an LLM is not written by "you", it was written by the people whose code the LLM was trained on, and thus it cannot be owned/copyrighted/MIT-licensed by "you". If this kind of code is allowed at all, then the project violates copyright and its license is meaningless.

So if I go read your code on github and then write a competing product based I what I learned then you own the copyright to my work?! Or if I read your book on programming, do you own the copyright on everything I write after that? No, sorry, that is complete nonsense and not even close to how copyright law actually works. It doesn't even apply to patent law, which is is something else entirely that we are not talking about here.

I am not “pro-AI” in any way, shape, or form, but your statement is just plain wrong, sorry. Please don't spread FUD here. Not for either side of the argument.

[jobukkit]

If you copy-paste code from someone else's GitHub repository, then yes, they still own that code and you cannot ignore its license.
If you read the code and then write new code yourself based on the knowledge you gained from that, then you do own it, because you wrote it based on your knowledge; the fact that your knowledge is based on other people's work is irrelevant.

LLM's do not have knowledge/worldview, they just regurgitate the text they were trained on. It's basically a faster, automated version of trying to code without a keyboard by copy-pasting snippets of other people's code, which would obviously be copyright infringement. We're nowhere close to real AI that can write original code.

I get that you're trying to be neutral but I don't see a lot of wriggle room for nuance, the vibe-coding situation is just grim.

[neurocyte]

I get that you're trying to be neutral but I don't see a lot of wriggle room for nuance, the vibe-coding situation is just grim.

I am not neutral (my opinions are way off in a different direction). I don't want political discussions here. That is what said. You are very welcome to come join our discord and have a feisty political discussion with me in the #offtopic channel.