FreeBSD's stance on generative AI

[rbranco]

My advice to young people is to avoid ruining their career prospects by listening to the wrong people that dismiss AI as something that will fade with a stock market crash, because it wil be quite the opposite... And those who talk about "enshittification" have a poor or delayed sense of smell. Where were they when GDPR popoops appeared?

AI can rewrite the Linux kernel in Rust. This article mentions 5 such private kernels:

See https://lwn.net/SubscriberLink/1099594/a52020921c2d8aef/

Some people may not like the ethics, but every tech has both its pros & cons, and AI is not different.

Lots of things are changing as we speak and people should avoid getting stuck in the complaining phase.

 

[T-Aoki]

My advice to young people is to avoid ruining their career prospects by listening to the wrong people that dismiss AI as something that will fade with a stock market crash, because it wil be quite the opposite... And those who talk about "enshittification" have a poor or delayed sense of smell. Where were they when GDPR popoops appeared?

AI can rewrite the Linux kernel in Rust. This article mentions 5 such private kernels:

See https://lwn.net/SubscriberLink/1099594/a52020921c2d8aef/

Some people may not like the ethics, but every tech has both its pros & cons, and AI is not different.

Lots of things are changing as we speak and people should avoid getting stuck in the complaining phase.

You SHOULD know that technical aspects and legal aspects are completely different thing.

From the view of technical aspect, I don't object to TRY AND LEARN AI/LLM for young students. It would be strongly wanted IN LATER YEARS.

But I object to merge AI/LLM-generated codes, as legal aspect of risks are NOT AT ALL cleared WORLDWIDE by international laws. So merging AI/LLM-generated codes without sufficient (would be quite a large task!) confirmations that the codes do NOT violating copyrights/copylefts/patents of incompatiblly licensed codes could make you criminal AT WORST. This risk SHALL be avoided. International law to clarify this is needed ASAP. I don't want anyone to become criminals.

These are different things. Try AI/LLM (if you ignore environmental aspects like too much energy consumptions), learn it, think how efficiently use it. But don't merge generated codes untill legal issues are clarified to be safe.

 

[gotnull]

Lots of things are changing as we speak and people should avoid getting stuck in the complaining phase.

One thing for sure the world we knew won't be the same.
I don't know how the AI tech will evolve but I agree with your statement, AI will not disappear, for me it's because of one single thing: money.
Too much money have been spent already, billions of billions, at this point it's too massive to get abandoned by any government or companies involved in that race, it's not about market anymore. Now is it good or bad? future will tell.

There is one thing that could stop AI though, and humans won't be able to control it, it's weather: water and dry are actually very concerning. That alone could stop everything (or at least reduce drastically) and it's already on march ...

To answer OP question, that's only my analyze and my opinion, the only reason why FreeBSD didn't say anything yet it's may be because they observe what others do.
I don't think they will position themself against AI usage, just under some conditions I guess. Frankly if they were against it they would have said it already.
At some point it's possible that most OSes will use AI to at least find bugs, write documentation, translation, AI looks like efficient for theses tasks.

 

[rbranco]

You SHOULD know that technical aspects and legal aspects are completely different thing.

We already know everything we need to know wrt the legal aspect. US courts, which are the only ones that matter, already dictamined that training is fair use.

And then also this: https://www.fenwick.com/insights/pu...t-holds-copilot-creates-new-code-not-stripped

• The U.S. Court of Appeals for the Ninth Circuit affirmed dismissal of claims under the Digital Millennium Copyright Act (DMCA) against GitHub, Microsoft, and OpenAI.
• The court found that generative AI tools such as GitHub Copilot and Codex, which learn from existing works but create new outputs, do not “remove” or “alter” copyright management information (CMI), such as author names, copyright notices, and license terms.
• The court recasts the “identicality” requirement as an evidentiary inference rather than a required element of the claim. Near-verbatim copying without CMI may still support the inference that the CMI was removed, while material differences suggest a new work that would never have had CMI attached in the first place.
• Plaintiffs cleared standing but lost the claim. The court rejected the “output” theory because Copilot, as plaintiffs described it, generates new works rather than copying existing protected works.

 

[sualbee]

I am trying to say, that ethics are not set in stone...

Surely they aren't, as history clearly shows us. So what? Do you think we should be concerned about ethics at all and, to the best of our abilities, try to live accordingly to the values we have now, or all of this doesn't matter? That was my question basically.

...beyond continuation of existence, let's say, the objective, ultimate ethics measurement.

Then what if we believed that relying on big tech LLMs only brings us closer to the end of FreeBSD (or personal sanity, or at least some IP protection, or Earth ecosystems - you name it)? Would that mean that such a thing should be considered objectively unethical as well?

Concretely, regarding FreeBSD, to survive the competition, it should not limit itself by arbitrary proclamations about the use of a technology.

Well, a lot of folks obviously don't agree with you on this for various reasons. I personally think it only hurts FreeBSD's competitiveness because, just as with its current relaxed LLM policy, the maintainers are still the bottleneck, and no one is able to land a lot of new features without going reckless either way. OTOH, "AI friendly" label by itself is a hard NO in certain use cases and for certain people, but maybe statistically it's not a concern at all.

Refuse to utilize gun powder, because whatever, and your lofty ethics will bleed out with you. Just don't store it next to your stove.

I think these bogus analogies don't really add anything meaningful to an argument, just like memes.

 

[rbranco]

One thing for sure the world we knew won't be the same.
I don't know how the AI tech will evolve but I agree with your statement, AI will not disappear, for me it's because of one single thing: money.

AI will not disappear for the same reason tech doesn't magically disappear. It's usefulness massively outweighs the cons.

Trains were like that at first. Then they got better.

 

[Beastie7]

They can afford to go Amish a bit more.

 

[T-Aoki]

We already know everything we need to know wrt the legal aspect. US courts, which are the only ones that matter, already dictamined that training is fair use.

And then also this: https://www.fenwick.com/insights/pu...t-holds-copilot-creates-new-code-not-stripped

• The U.S. Court of Appeals for the Ninth Circuit affirmed dismissal of claims under the Digital Millennium Copyright Act (DMCA) against GitHub, Microsoft, and OpenAI.
• The court found that generative AI tools such as GitHub Copilot and Codex, which learn from existing works but create new outputs, do not “remove” or “alter” copyright management information (CMI), such as author names, copyright notices, and license terms.
• The court recasts the “identicality” requirement as an evidentiary inference rather than a required element of the claim. Near-verbatim copying without CMI may still support the inference that the CMI was removed, while material differences suggest a new work that would never have had CMI attached in the first place.
• Plaintiffs cleared standing but lost the claim. The court rejected the “output” theory because Copilot, as plaintiffs described it, generates new works rather than copying existing protected works.

If you're a US resident (I don't know whether or not), it COULD be sufficient for YOURSELF ALONE, I agree.

But FreeBSD is used all over the world. So if any country still has risk and LLM used any code from there to learn, there can STILL be a serious risk when used at least in the country. So without international laws, covering and approved all over the world is mandatory to be safe. Safe in some (but not all) countries is NOT AT ALL sufficient to be really safe.

 

[rbranco]

If you're a US resident (I don't know whether or not), it COULD be sufficient for YOURSELF ALONE, I agree.

But FreeBSD is used all over the world. So if any country still has risk and LLM used any code from there to learn, there can STILL be a serious risk when used at least in the country. So without international laws, covering and approved all over the world is mandatory to be safe. Safe in some (but not all) countries is NOT AT ALL sufficient to be really safe.

I reside in the EU but the EU is irrelevant on everything that matters in FreeBSD and FOSS at large. US courts decide everything.

 

[T-Aoki]

I reside in the EU but the EU is irrelevant on everything that matters in FreeBSD and FOSS at large. US courts decide everything.

Cannot fully understand.
Copyright holders in outside US can sue at where the person / organization belongs to. Things are often not that simple with regard to legal issues.

 

[DanielBowman]

You SHOULD know that technical aspects and legal aspects are completely different thing.

From the view of technical aspect, I don't object to TRY AND LEARN AI/LLM for young students. It would be strongly wanted IN LATER YEARS.

But I object to merge AI/LLM-generated codes, as legal aspect of risks are NOT AT ALL cleared WORLDWIDE by international laws. So merging AI/LLM-generated codes without sufficient (would be quite a large task!) confirmations that the codes do NOT violating copyrights/copylefts/patents of incompatiblly licensed codes could make you criminal AT WORST. This risk SHALL be avoided. International law to clarify this is needed ASAP. I don't want anyone to become criminals.

These are different things. Try AI/LLM (if you ignore environmental aspects like too much energy consumptions), learn it, think how efficiently use it. But don't merge generated codes untill legal issues are clarified to be safe.

Laws are essentially letters to Santa Clause.
International law is misnomer. Basically the law of the jungle as there is no one impartial enforcer.

Yes, there are risks someone will try to kill and or extract resources from FreeBSD foundation based on copyright claims via Ai implementation. But
a] no one, who wants to remain relevant, can avoid it
b] FreeBSD is not exactly the prime target for extortion, being foundation, open source and free
c] gains vs risks in favor of Ai in FreeBSD as it could never afford army of coders like other projects

 

[CShell]

Laws are essentially letters to Santa Clause.

Good luck with that

 

[DanielBowman]

Surely they aren't, as history clearly shows us. So what? Do you think we should be concerned about ethics at all and, to the best of our abilities, try to live accordingly to the values we have now, or all of this doesn't matter? That was my question basically.

Then what if we believed that relying on big tech LLMs only brings us closer to the end of FreeBSD (or personal sanity, or at least some IP protection, or Earth ecosystems - you name it)? Would that mean that such a thing should be considered objectively unethical as well?

Well, a lot of folks obviously don't agree with you on this for various reasons. I personally think it only hurts FreeBSD's competitiveness because, just as with its current relaxed LLM policy, the maintainers are still the bottleneck, and no one is able to land a lot of new features without going reckless either way. OTOH, "AI friendly" label by itself is a hard NO in certain use cases and for certain people, but maybe statistically it's not a concern at all.

I think these bogus analogies don't really add anything meaningful to an argument, just like memes.

Yes. We should be concerned about ethics and morals. It is about our survival as individuals, families, tribes, nations, species and life itself. But Ai is not clear cut case of right or wrong. Like unnecessary cruelty would be, for example. It is a technology with many uses. Banning it is irrelevant, for project like this anyway.

Pros and cons regarding monopolization, IP, ecosystems etc. are not clear cut. They have mitigating pros and are bound to evolve.

Yes, we disagree. On the competitiveness as well. Time will tell. In the mean time I prefer for options to be open. Let's see and judge on the merit - will it help to churn out better security or wifi drivers or whatnot? If I lose contributions from people who do not like options open, I wonder why they are in FOSS...

All analogies are bogus. This one was rather good one, tho - to remind about the difference between technology in general and it's particular use.

 

[JohnK]

I think it's disgusting that all these ai marketing people/posts have inundated these forums built for people.

////

 

[NinjaBSD]

Humans are inherently lazy. For the most part the vast majority of people will take the easy route (all-else being equal or if only the outcome is the status quo).

The issue with AI used within FreeBSD code base (or any software, open source or not) is that as it starts being used for additional features the human aspect will start to fade as time goes on. Sure, at the beginning, it may have already been used to assist with some features and the maintainer manually reviews over before committing. However, as time goes on the "I've only found a few potential issues AI has given" attitude translates into a bit more compliant (or complacent or laziness) and starts accepting AI output without much if any manual review. This is a major potential issue; not necessarily in the short term but in the medium and long term. As AI becomes more and more accepted in day to day workflow the human knowledge levels will decrease and reliance on AI will increase. Think about the youngsters today and their lack of fundamental computer knowledge. They think they have it easy because they don't have to know how a computer works to operate it but the issue is medium and long term the knowledge transfer hasn't and wont be taking place and so then we have whole civilization that basically doesn't know how anything *actually* works. So sad but maybe this is just what happens when one gets older and starts saying "kids these days".

Since FreeBSD is open source there is no issues with leveraging AI for certain aspects such as security/vulnerability scanning and such but in terms of adding features, programs, etc my take is it is a slippery slope and once the cat is out of the bag it will be impossible to put it back in. Either way, I am confident the core team will not make quick rash decisions so I'm happy with whatever the outcome.

 

[Phishfry]

AI will not disappear for the same reason tech doesn't magically disappear. It's usefulness massively outweighs the cons.

Trains were like that at first. Then they got better.

Its a matter of semantics. Trains were first called locomotives. Then transformed by electricity to trains.

AI is some blase terminology that really doesn't seem to fit what it does.

I see the term Machine Language tossed around and that really seems to be what we are talking about.

Train a computer to sort though a dataset of stars and find anomalies.
Somehow everything is AI now.

 

[mer]

So we have licenses that Linux will accept and licenses that they won't.
We have BSD licenses and licenses that are compatible with BSD licenses.

If AI is trolling/consuming everything from all different licenses, then what?
If Linus states "Linux will accept AI generated stuff" does that mean if AI generates ZFS stuff, Linux will accept it?

What happens if AI generates code that sourced from GPL, CDDL, MIT, BSD licenses. What takes precedence?

Phishfry I think is touching on it. What exactly is AI, is everything AI?
As for Jessies jacket, I'd start with calling 867-5309

 

[DanielBowman]

Its a matter of semantics. Trains were first called locomotives. Then transformed by electricity to trains.

AI is some blase terminology that really doesn't seem to fit what it does.

I see the term Machine Language tossed around and that really seems to be what we are talking about.

Train a computer to sort though a dataset of stars and find anomalies.
Somehow everything is AI now.
If a missile can spot the difference between a pitchfork and a gun is that AI?
Because that capability was around long before AI. But now rebranded. Super Intelligence....

Imagine when this bubble pops. It's already sinking the bond market with junk level commodities.
GPU in datcenter has how long of a lifespan??? Nobody knows for sure do they.
Write off how much for Jessies new jacket?

You might be confounding founding companies with the technology. Their market value might (most likely will) crash, given their "business" model. Technology of machine learning via transformers will not crash. Even the LLM chatbot subset of Ai /.. Machine Learning. Let alone other fields - driving, protein folding etc.

 

[bakul]

Compiler for a recent programming language where the primary author uses AI heavily:

$ git diff HEAD HEAD~10|wc
    6229   32309  290068
$ git diff HEAD HEAD~100|wc
   90618  408168 3220667

FreeBSD main branch:

$ git diff HEAD HEAD~10|wc
     488    2073   15873
$ git diff HEAD HEAD~100|wc
   12433   49063  397010

Measuring diffs over 10 and 100 commits.

This is a crude measurement but it demonstrates one of the problems: AI written code seems to have a lot more churn. In case of this PL, self compiling performance has varied by a couple of orders of magnitude and fixed things break again. Over and over. I wonder how much of the AI written code will remain maintainable over a long period of time.

 

[sualbee]

Yes. We should be concerned about ethics and morals. [...]

If that's the case, your remark about a reality check and ethics transience was pointless. The OP's concerns are reasonable and valid. You think the benefits, whatever they might be, of using and promoting commercial big tech LLMs outweigh the price everyone is paying. Someone else might think that's very far from being the case. But the thing is that we all live in the same reality, the difference is simply in our values.

 

[rbranco]

Its a matter of semantics. Trains were first called locomotives. Then transformed by electricity to trains.

AI is some blase terminology that really doesn't seem to fit what it does.

I see the term Machine Language tossed around and that really seems to be what we are talking about.

Train a computer to sort though a dataset of stars and find anomalies.
Somehow everything is AI now.

After passing the Turing test there's really no point in moving the goalposts. Where I previously used "LLM" and double-quotes around AI, I no longer have such taboo now.

And the train analogy was to highlight that they were heavy pollutants at first and gradually got better. The same with happen with everything around AI.

 

[bakul]

Its a matter of semantics. Trains were first called locomotives. Then transformed by electricity to trains.

The word train for a series of connected carriages drawn by some power was used long before even the steam locomotives were invented. AFAIK the word locomotive (in the sense of having power to move by itself - by 1814) was used for engines only.

 

[Phishfry]

Even the LLM chatbot subset of Ai /.. Machine Learning. Let alone other fields - driving, protein folding etc

The Human Genome Project is going to revolutionize healthcare.
Oh wait that was watson....
So much fluff these days in the IT world it is incredible.

some will call this circular financing

Who’s Really Paying for the AI Boom? The Bond Market Is Starting to Ask

Bond investors are quietly repricing the debt that funds AI's massive buildout, and some of the numbers coming out of quarterly filings suggest the easy money phase may already be over.

finance.yahoo.com

Nvidia has partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize

If there ever was an AXIS of EVIL that is pretty much it.

 

[DanielBowman]

If that's the case, your remark about a reality check and ethics transience was pointless. The OP's concerns are reasonable and valid. You think the benefits, whatever they might be, of using and promoting commercial big tech LLMs outweigh the price everyone is paying. Someone else might think that's very far from being the case. But the thing is that we all live in the same reality, the difference is simply in our values.

Nice strawman you have there.
a] Ethics of a gun is pointless. Ethics of unnecessarily shooting people is paramount.
b] Promoting commercial big tech AT&T Unix systems, by advocating for not banning the use of computers....hmmm

But we can agree we have different values. I for example value the truth more than a fear. Therefore I prefer options to be open.

 

[DanielBowman]

The Human Genome Project is going to revolutionize healthcare.
Oh wait that was watson....
So much fluff these days in the IT world it is incredible.

Who’s Really Paying for the AI Boom? The Bond Market Is Starting to Ask

Bond investors are quietly repricing the debt that funds AI's massive buildout, and some of the numbers coming out of quarterly filings suggest the easy money phase may already be over.

finance.yahoo.com

If there ever was an AXIS of EVIL that is pretty much it.

Machine protein folding is already done deal. Few years instead of million years. Nobel prizes collected, public database filled, major hurdle for research gone.
So is driving. Pretty much solved.

Financial sector concepts from circa 1913 are atrocious ( I call it the scam of century).
But again, that does not void value of the machine learning transformer technology. It is just over-valued regarding existing companies operating it.