What are the AI usage guidelines for core OS contributions?

Development OS
1

I know. However, saying that

very much implies that there was some intention to send this upstream or somehow (re)open a discussion because there is some intriguing work that cannot be merged as it breaks contribution guidelines (Especially if claude tells you this, however, I don’t want to accuse people of actually being malicious and I assume that that has been missed). And I don’t understand why there would be progress reports here if the goal would not be to ever try to upstream this.

It cannot be reimplemented. It would still be with AI assistance which still would be against the contribution guidelines as far as I understand. That’s part of the reason of why I don’t understand what the end goal of this is supposed to be. Even if the end result is stable enough and cleaned up well enough, it could not be merged. The only thing that I see is a potential source for patches that mustn’t be upstreamed, potentially making it even more difficult to work on the actual powerpc port. Even if omgmosters port lives as a separate distribution, I don’t see how it would properly interact with the rest of Haiku.

I have made some progress on the PowerPC port
2

It need clarification. If any kind of AI output can’t be used as reference to make Haiku contribution, than it makes no sense. For example if AI found real human verified bug, bug will be unable to be fixed because AI has been used as reference. Or if AI has been used in some programming articles that human read to get programming skills and then used to contribute to Haiku. I do not think that it is realistic to fully disallow any kind of indirect use of AI tools because developers may be exposed to AI tools output even if they are not aware of. And a lot of information on internet/books etc. is not fully traceable to all its origins.

If any AI references will be disallowed in strongest possible form, than all Haiku developers must live in some “no AI village” to avoid to be tainted by being exposed to some AI output.

9 Likes
3

Exactly! These days even googling documentation exposes you to AI. Dodging AI generated stuff is beginning to remind me of some bullet hell games…

4

If you use “documentation” generated with LLMs to directly base your changes on, this is AI assistance. I think that’s quite clear in this case. And omgmonsters is the person that’d be in control to (not) use it

5

I do not think it is that clear.
There is a difference between

  • writing directly code with AI where most people essentially do not control anything regarding quality/resusability/copyright
  • and a human reading some project doc (even if generated by an AI) to write new code based on that documentation.

In the second case you are actually protected by law (at least in the EU) as you are free to write any code from scratch to interoperate with the HW you own. That does not negate the negative impacts of LLM, but you are at least writing the code from scratch.

I interpret the ban on IA assistance as one on directly writing with LLM, as the second point is not enforceable anyway: even somebody willingly trying to avoid AI generated documentation may use some without knowing. And good luck on proving submissions are written based on LLM generated documentation.

Edit: this debate should be its own discussion so we can stay on the original topic

2 Likes
6

This is just your opinion. The official Haiku policies are located at Policies | Haiku Project, and not in AGENTS.md (which is a file that is supposed to be read by LLM agents, and not humans).

The Haiku project cannot accept contributions (e.g. code, documentation, translations, etc.) which are under ambiguous or incompatible licenses. (This includes most content produced with the use of LLMs.)

If you don’t agree with that definition or feel like it needs to be clarified with Haiku maintainers feel free to start a separate topic.

4 Likes
7

I strongly agree with this actually.

I am not going to send any code from this upstream. I am also not going to submit any hand-written code from this project, because I believe that by the nature of this project, even hand-written code is strongly influenced by other work that has had AI input or generation.

I’ve decided not to include a detailed checklist of changes and work implemented to bring this up on real hardware for hand-reimplementation in Haiku. That would really seem to go against the spirit of Haiku’s policy since the majority of the work and especially debugging was done with AI assistance.

I’m planning on releasing this with it’s own branding in accordance with Haiku’s policy on distributions.

I love the Haiku project and I’ve been a huge fan and user for many, many years. I would not want to do something that would compromise the project.

At the same time, I am really proud of the work I’ve accomplished here, and I hope the very few people who still have working Apple PPC hardware might find it interesting :slight_smile:

4 Likes
8

I worry about this a lot actually. Back when I was writing code in my day job pre-AI, I was very often copy/pasting from Stack Overflow and then editing when I couldn’t figure something out. I have a sneaking suspicion that much of what is currently submitted to SO has AI influence, so I think I can’t even do that any more if I want to fully avoid AI influence.

I think AI influence is already everywhere… :grimacing:

9

Wait… We ban AI generated code because it may not match our style, we never know where AI took this code from that it maybe a GPL or even leaked proprietary code. We dislike how AI do things in redundant way. If it is a clean room reimplementation given by results of AI research (for what to do but not how to do), do these problems still exist?

10

I don’t think it’s as black or white as people make it out to be. Yes, Haiku will never upstream your vibe coded project. But agentic engineering is a lot different than vibe coding. Haiku will not directly upstream that either, and that is ok. There is definitely interest in clean room AI implementations within the main Haiku dev community. The main thing with submitting AI assisted upstream requests is that this stage needs intelligent, skilled, and human responses. Not even Claude will get this layer right atm. Telling our agents not to make upstream requests is the easy part. We also need to tell our agents to always pay attention to licenses, to never copy GPL or proprietary code, and when that is all said and done, it also helps if we run all code through free code audits to ensure we are not directly copying code from GPL or proprietary sources. Haiku is being extremely cautious, and I get it. Linux has fully embraced AI help, but Tovalds himself probably has more lawyers on retainer than Haiku could afford in a year. Still think they are a bit overly cautious. But I get it. Without court precedents, it’s hard to say what outcomes might be. I just think that AI is “too big to fail”. Meaning that vibe coded projects may not pass the copyright scrutiny test, but any real documented effort of human software engineering will most likely be upheld.

The interesting thing is that the part of my project that Haiku devs are most interested in atm is the MIDI drivers, which, while no GPL code was used, technical information was very reliant on ALSA documentation for USB MIDI quirks. This has traditionally always been legal. But my USB Audio changes which were all reverse engineered, using Wire Shark and other traditional methods, have not gained as much interest yet.

Just trying to emphasize that we are still in the early days of this new dynamic. We will figure it out.

2 Likes
11

The single reason that was retained for banning LLM generated code in Haiku is copyright concerns, because this one was easier to get consensus on. But, indeed, that only makes sense if the code is directly generated by the LLM. If you use it as a study tool, and to generate proof of concepts, then I don’t see how that rule can stop your code from being accepted.

As much as I (and other developers) would prefer a stronger ban for different reasons (climate impact, direct impact of the scrapper bots hammering our servers and increasing our costs and time spent dealing with that instead of writing code, low-paid people doing the classification of training data, getting people to use closed source tools and paying a subscription so they can code, etc), these reasons did not get a consensus and were not included in the current rule. Some people successfully argued that these things are OK or don’t matter.

There was not much debate on wether the quality of the contributions would be acceptable or not: experience from other projects show that there will be a lot of low-quality work, some of them (like Godot) already having changed their mind on accepting LLM generated code because of that. Some of the existing developers have confirmed that in the current state of things, there is no chance that LLM generated code would meet Haiku’s high standards for getting things included, and human interaction would be needed. However, this may very well change in the future, and we already have rules in place to reject code that doesn’t meet our quality standards.

So, what’s the outcome? Technically, contributions made with the help of an LLM but where the code was written by a human are currently allowed. However, you need to find someone willing to review it, and some of the most active developers in Haiku will very much not be interested in doing that. So it may take some time until someone does take up the review, or it may never happen. Or we may just decide to reject it anyways, but that will surely will cause some more drama.

13 Likes
12

I share a lot of @PulkoMandy’s concerns but I’d like to bring up one thing: getting older hardware to be supported is a real challenge. I’d bet there isn’t a lot of overlap between the people who know how to program the older systems and the ones who wish to keep the older system running. Low-level bits for boot-loaders and hardware-specific needs for an OS (and other projects) for old hardware is a really niche thing. Besides Haiku, I’ve seen other LLM work to get things like modern Firefox and other apps running on Classic Mac OS and PowerPC Macs. Sure, there are some LLM concerns, but the end result could mean more reasons for people to hold on to those systems and spare them from being sent to the dump to become e-waste. Maybe that tips the scales a little bit when it comes to economical and environmental concerns.

Otherwise, I’m amazed as the progress done recently to get Haiku booting on PowerPC (and ARM)! This is awesome and I hope that the code can be contributed. Perhaps with the help of others to ensure that a clean git history is maintained, that the code is maintainable and intelligible.

Maybe I’m just silly but it would be neat to have some additional platforms and re-visit the Fat ELF discussion.

1 Like
13

When my hobby becomes a thing that kills the planet I live in, I don’t shrug and say “well, that’s fine, now I have robot friends to play with”. I will find other things to do.

These systems will not have practical use. The combination of legacy host software and LLM generated code is a security nightmare. The nostalgia for systems that are simpler and easier to understand is gone if you don’t do the understanding yourself. The ecological/permacomputing aspect is gone, because the software production costs a lot more, and the computing per watt of these machines is hilariously bad. What’s the point? It’s disgusting and ruining the thing it pretends to keep alive, by completely misunderstanding what it’s about.

And in the process, as you say, it prevents people from growing new skills. Which were already out there on the searchaple internet if you cared to look for them, that’s how they entered the training data. Or you could have asked this and other forums for help. Met some people. Worked together to figure it out. You know, the things hobbies are good for?

3 Likes
14

This is a really great discussion. I am a software engineer based in South Korea, and recently—with the help of Claude Code—I managed to install Haiku and Tailscale on my i386 machine. Since I can read kernel code to a certain extent, I verified the bugs that Claude 3.5 Sonnet found and reviewed and approved the code it generated. Knowing the community’s AI ban policy beforehand, I had no intention of submitting it to the official Haiku codebase; I only shared it informally.

This is certainly a complex issue. Where I work in Korea—and more broadly across East Asia—the tech culture is heavily VC-driven. As a result, almost all developers are pushed to use AI without much deep discussion about the fundamental implications of AI-assisted contributions. – In fact, a famous Korea-Japan-based messaging company even evaluates developers’ technical performance based on their AI token usage. :slight_smile: – Because of this background, I am really glad to see this discussion happening and hope this thread stays active.

That said, as a long-time fan of BeOS and Haiku, I do have a slightly different perspective on whether a total ban on AI is the only right path forward. (Though I want to re-emphasize that I fully respect the Haiku community’s decision.) For example, AI could potentially discover solutions for hardware bugs—whether they are limited to specific devices or affect hardware universally—or it could help resolve long-standing issues that have been buried in TODO comments. Of course, I completely understand that maintainers cannot possibly manage a massive flood of these proposals. (I experience the exact same struggle at my own workplace.) While I agree that mindless “vibe-coding” creates a terrible ecosystem, I also wonder if projects that rely heavily on open-source and voluntary participation might miss out on valuable opportunities by rejecting AI entirely.

I hope my intentions come across clearly and constructively. To emphasize once again, I find this discussion immensely valuable. Please don’t ban me! :slight_smile:

8 Likes
15

Please do not start that again, really. Data centers running LLMs working no matter of our hobbies. There are much bigger players involved.

5 Likes
16

Yes, of course, and the same applies to travelling by airplane or even by car. Yet, I don’t do these things either, even if my own choice will not change anything directly.

What will change things is that I still live a pretty normal life, and I even have cool stories to tell about crossing France by bike or Europe by train. So no one around me can say “living without a car? No one seriously do that!”. So it is for LLMs. I will do my best to show that yes, building software without LLMs is possible, it works, it has lower impact on resource usage. And I talk about it if I want to. If you are not happy with that, report me to the moderation team until I get banned from this forum. I do not intend to shut up about this.

5 Likes
17

This is the best comment on the topic that I have read so far. Thank you @rainygirl !

18

|| ||
|| DISCLAMER : ||
|| This post is quite long and if you feel yourself overwhelmed when you read ||
|| such long texts, then just watch the videos - if you want or they made you curious - ||
|| and skip my written content. ||
|| This post is combine several answers, and ||
|| in some cases, those rather confirmations of concerns, ||
|| and some cases agree on opinions cunducted above, ||
|| meanwhile show the reasons which makes them to understand better ||
|| both their existence and why they can be real and rightous revelations. ||
|| ||
|| Thank you in advance for your understanding. ||
|| ||

Recently I saw a video from ActionRetro channel about an OS and a Github repo that contains a similar - but far from meticulously precise as Haiku did to re-create -, but they rather revise-engineer it on the look and feel (tha’s my opinion) than functions in the deep. So similar effort that Haiku targeted BeOS R5.
They targeted MacOS 7 - but as the video progressed, for me, the result concluded me in my opinion above. This is the way concers from one side Haiku developers from using AI.

https://www.youtube.com/watch?v=rJRlHKQqX2M – Mac OS 7… on x86?!

And also I have another video, that support your your statement that it may be necessary to use computers that are already older, but still very usable, in the present
→ on the one hand, due to the intentions of mainly Western global companies, with which they want to force people into subscription schemes, saying out honestly : to eliminate truly the personal computers you own personally and use persionally regarding to your interests.
Exactly what caused the explosive development of the IT industry, the accumulation of their capital, but all of this can no longer ensure the correctness of the decisions made in the present, the future of the business, the dividends to be paid to their shareholders, because an unrecognized direction or development can deprive them of a continuous or increasing profit rate. In this way, they want to secure themselves, they want to secure subscribers, so that even if they do not develop or only sell a smaller number of new products, the customers will continue to keep the old ones, because their own lives have become stricter, the current profit, or at least the income, should be secured - above all.

https://www.youtube.com/watch?v=VEiOV6JL2bg – The Storage ECONOMY Is Broken. MicroSD Is OBLITERATED.
The first wave of these was cloud solutions - fortunately, you can now implement them at home, but the masses will not strive to implement them, because it costs money and acquiring new skills, and the companies’ are there, and they started selling them cheaply by feeding them. Then, hip-hop, your personal data is no longer yours. There is no memory card in your phone, because send it to the cloud.
The second wave is AI, which could be used to make life easier, to advance science, and to improve people’s living conditions, but it is being developed into an addictive service that is presented cheaply and then operates with rising subscription models. Yes, you can install this yourself, on affordable hardware at home, but for now, their power is not as amazing because it can run on smaller hardware and allows for coding, vibe coding, rather than image generation or video generation.
This AI boom, in itself, has brought about an astronomical increase in PC parts and components, because there is a given available production capacity, and only the entry of China can give hope, because there is a skill that can shorten the construction of a factory and the establishment of a production line to a much shorter period than can be observed in the case of an increase in production capacity implemented in any region of the Western Hemisphere.
This is possible because previously server farms were assembled from a large number of cheap PC components, so home users and companies themselves had to share the available manufacturing capacity.
Then in the meantime, Nvidia started producing ASICs for crypto miners, for video games, design, and to replace video cards built into scientific and government servers, and also started producing other target hardware that was needed for server farms for cloud services and AI server farms, but it also ordered large quantities of high-bandwidth RAM and storage, and other non-manufacturers, but also service providers, such as Google, Amazon, Microsoft, Tesla etc. ordered chips, memories, storages to build additional server farms.
This way old PCs (even 10-14 years old ones) were started to worth keepin’ for home users, and 4-10 years old for firms as well. Too expensive to follow new releases, not everything requires bleedin edge harware.
This way start to be emerging the re-cycling PCs, tablets, mobilephones and go tosale or some cases give them away to those who in need (no many to buy new one or buy any at all). To keep them in the information level society - to maintain their necessity to be keep up the connection capabilities that a necessity nowadays, and also keep on rational cost rates or more affordable maintenance price range than buying a new device.

https://www.youtube.com/watch?v=Mo98p0ziJvQ – They’re Taking Away Your PC

I terribly sorry if this post better fit to a blog expand than a forum post, but once was asked me to answer rather in one post on the forum than several ones - so I just follod this advice in this post.
I wrote a disclamer finally.

19

Unfortunately I cannot fixed some typos, syntax errors or completed sentences for better understanding in my previous post - I hope it will be clear against all.
The 3rd video about reason the keeping older PCs, the 2nd video is confirmation of a trend to glue us into subscriptions.

20

This is how AI bubble comes.