Will FreeBSD adopt a No-AI policy or such?

[DanielBowman]

These campaigns are often just

Blacksmiths against CNC machines.

Vibe coding is different matter. It is just pushing random buttons on CNC machine until something looking useful comes out.
If someone is willing to stake their name and reputation on it, his fault.

Btw you will definitely need AI to at least filter out the mountains of AI slop commit requests in open source projects. To do that AI needs to have very good baseline. So you are basically paying AI to train it for free.

 

[itstotallyme]

This means little.
What people want to fight against is "vibe coded" software or PRs.

This is not representative of all voices in this discussion.

There are good reasons to ban any LLM output from getting into the project, as did NetBSD, SDL, postmarketOS, many others.

A ban is also useful even if it doesn't 100% stop people. Can you name a ban that 100% stops people? Do you have a code of conduct, does that 100% stop people? It's still a useful signal.

Anyway, I understand people here have a different opinion. I just wanted to clarify that the discussion is more complex. Even though it seems like FreeBSD will apparently remain relatively pro-AI, at the end of the day (I'm saying this compared to the position I have, not as an absolute measure). I'm thankful for the clarity, if anything.

 

[LibreQuest]

This is not representative of all voices in this discussion.

There are good reasons to ban any LLM output from getting into the project, as did NetBSD, SDL, postmarketOS, many others.

A ban is also useful even if it doesn't 100% stop people. Can you name a ban that 100% stops people? Do you have a code of conduct, does that 100% stop people? It's still a useful signal.

Anyway, I understand people here have a different opinion. I just wanted to clarify that the discussion is more complex. Even though it seems like FreeBSD will apparently remain relatively pro-AI, at the end of the day (I'm saying this compared to the position I have, not as an absolute measure). I'm thankful for the clarity, if anything.

True. I don't understand the push for the use of proprietary systems in development of open source. Makes about as much sense to me as devs pushing the use of other proprietary software or systems in development. But in this case the software injects licensed code into projects. So not only are the systems themselves proprietary but their output is borrowed licensed code not to mention flawed and potentially malicious.

I like to change AI for iTunes. Makes the mantras more fun to read. "ITunes is just a tool" "you can't put itunes back in the bottle". Well, now I guess I have to use itunes it's the unavoidable future, don't wanna get left behind.

 

[malco_2001]

I can respect projects that do not adopt AI such as NetBSD. I do see both sides of it. With NextBSD in only a short time I have been able to:

* Port all the popular KMS GPU drivers to kernel extensions with plists the kernel will autoload via IOKit (There is no configuring, graphics just work, even virtual machines now using KMS which to my knowledge no other BSD has at this time) AMD/RADEON, INTEL, NVIDIA, ported more drm-kmod drivers as kexts for VBOX, BOCHS, VIRTIO - DONE
* Make UTM with graphics acceleration (modesetting, not SCFB) and sound fully supported on Apple Silicon Mac - no other BSD can do this - DONE
* Boot the RPI 500+ 16g ram model natively (no UEFI) with video (PRs landing this week to finish the image) - again no other BSD can do to my knowledge - IN PROGRESS

At least a few engineers I can think of who make projects I like and use detest AI. I read their comments, statements, and I wonder can they just not see what I see? Just look at that list. On the other hand I could see it being annoying to review a lot of bad pull requests. I would just write a bot to close them all. Even if I had more experience than I do, which I believe is not nothing, I still would use AI to accelerate my efforts, hands down.

 

[msplsh]

I wonder can they just not see what I see?

Externalities.

 

[astyle]

Well, if you use AI as a tool, rather than look at it as brains, it can help. Even Linus Torvalds used AI to help him debug an Intel Xe GPU driver - and FreeBSD is gonna benefit from that. Even KDE Wayland on FreeBSD had a bit of help from ChatGPT - there was a patch for x11/kwin that was an absolutely crucial fix for a showstopper bug, a story you can find in this
Thread trying-to-run-kde-6-plasma-with-wayland.93951, before page 30.

Blind faith in AI is something I would personally discourage. "Productivity" is NOT "How many patches per hour can we generate and submit". Yeah, it's nice to see a problem actually solved sooner rather than later.

But, my thinking goes, maybe there should be some guardrails against submission of 'bad pull requests'. Like submitting some evidence of testing that would be difficult to dismiss as "AI hallucination".

After all, there's kind of a reason why devs of the Linux networking stack decided to shut down submission of AI-generated patches.

 

[bongo_x]

Well shit. I had stupidly assumed FreeBSD would be anti AI just based on the overall philosophy, and have been playing around with it and putting my efforts there. I should never assume, people will always disappoint.