Software Freedom Conservancy

@conservancy LOL WTF NOPE.

"FOSS projects should not shun contributors who choose to use LLM-gen-AI systems."

🤡

@dalias @conservancy kinda stopped reading after

“There are many intersecting ethical and moral issues regarding these systems, many of which are not currently fully understood.”

They are well understood and visible.

@conservancy I'm going to state this very clearly:

Projects/communities have an absolute right to exclude people whose behaviors they deem unethical, unsafe, or hostile from participation.

And deeming LLM-gen-AI use as unethical is a completely reasonable position.

@conservancy Ok. With that statement, what exactly is SFC "conserving"? The right to use, and not be ridiculed for using, unethical slop in FOSS?

@conservancy

There is less said on the BitKeepering of project tooling wrt process; the difference between toleration, blessing (i.e. tutorials) and requiring; the use of these tools to do reviews themselves; and the nature of using them for social interactions in exchanges, and a special treatment for english-washing for people embarrassed by their English proficiency. And obviously reputational harms.

But it's a step in clarity on the SFC's position. If a touch narrow for me personally.

Yeah. The most generous reading would be that they mean "don't bully people or act like a jerk, just gently close their contribution and invite them to contribute in other ways"

Though I did also read it the other way too.

As for "are not currently fully understood" this is code for "we can't say what they are without offending some of our projects" a diplomatic retreat from having a position or making a conclusion.

@rotnroll666 @dalias @conservancy

@conservancy encouraging and supportimg LLM slop is contrary to the very idea of free software

@doctormo @dalias @conservancy 👀 indeed.

The rest is not so bad tbf.

@dalias @conservancy oh yes they should

@conservancy you are just 🤡

@rotnroll666 @doctormo @conservancy It's way too much weasel language, concessions to AI-slop interests, and phrasing that can be weaponized by the slop proponents against us for DARVO antics. A good version of this document would have been less than 10% of the length and valued people who create not industry and abuser interests.

@dalias @conservancy I still remember the exchange you had with that UX guy that criticized Fedi for not gaining traction, and then questioned *your* *personal* power to judge if something is right or wrong. These people cannot envision a world in which a community coalesces around a limit they deem important, and fully rejects something. They don't respect agency, unless it's a corporation's, in which case they just let it do whatever they want, kind of an extension of the "they must be right if they got all that moneu and power" kind of thinking.

@conservancy i would really like to see a stronger stance against these dehumanization machines, honestly..

@dalias @conservancy I'm all for shunning such "contributors" even harder. And ridiculing, there should probably also be some ridiculing.

@ticho @dalias @conservancy Fingerpointing.

@jens @dalias @conservancy I was thinking more something along the lines of The Simpsons' Nelson Muntz, but yours works too.

@conservancy LLMs are fundamentally incompatible with free software. This should be your entire policy statement.

@conservancy "thoughtful and ethical use of LLM tooling" is a double contradiction, this is disappointing

@conservancy Do you also recommend that people eat rocks? Do you have any guidelines for putting rocks on pizza?

@conservancy point and laugh, children. point and laugh.

@dalias @conservancy I do understand your point that the way LLMs were sourced and their power dynamics are at least problematic. Still there might come a time when we all will struggle to identify LLM code. This is the point where bad code, written by humans, might be confused with LLM code and real humans get blamed for using LLMs. I think realistically it well be more helpful to disallow or discourage huge PRs and bad code. I am not happy about that but it seems more realistic to me.

@chris @conservancy Um, fuck no. Just because we can't 100% reliably identify hostile and unethical behavior doesn't mean we just say "doing that is fine". The statement of values is the point. If there are shitty people who get a kick out of violating our consent, they will demonstrate themselves that in other ways too, and we will remove them.

@conservancy let’s make this perfectly clear.

Fuck all the way off.

And when you get there fuck off some more.

The less-nuanced version of this statement is still in preparation.

@dalias @conservancy Don't misunderstand me. I don't have anything against disallowing AI in your project.I just think that it might be like fighting windmills. I am fairly certain that I can create an LLM assisted PR that you will not be able to detect because I can actually write good code.

I just think it's sad if we started mistrusting each other over LLMs. It's okay to ban commits clearly marked as LLM generated. If not I'd rather assume it's a human.

@dalias @conservancy @wyatt honestly define "shun" tbh, because according to ai bros having a policy of lije " please dont use this here " is shunning them, when typically that is not what that would mean, im inclined more to think along the lines of .. "your bad for using this" or .. "if you use this then your ..." when i hear that; which is something else

@wyatt @conservancy @Li I will "shun" them in the sense of recommending against using (especially depending on) other projects by them or in which they are heavily involved, speaking publicly about/against their attempts to normalize "AI" use, etc.

I will deem hall-of-shame/shitlists of slop coders as legitimate and non-abusive, and will happily share them.

If someone thinks any of this is "poor conduct"/a violation of their principles of how we should engage with "AI"/a violation of their code of conduct, they can get fucked.

@azonenberg

On the contrary - if the indications hold that the output from LLMs are not copyrightable that means that LLM produced software has the equivalence of the Creative Commons Zero license, e.g. public domain.

That's how I started releasing software I wrote back in the 80s and I most definitely consider that to have been free software.

@conservancy

@troed @azonenberg @conservancy Surely you realize the copyright issue isn't the only one at play here?

@dalias @wyatt @conservancy im not sure what i actually think about this, so i wont comment much on this

i was just pointing out theres two completey different meanings here;

i would like to ig mention that having a list of "bad coders" has an issue of making it difficult to be accepted back if they stop using them.. which is presumably what you would want them to do .. (this is an issue with blocklists in general though.. tbh)

im not sure what i think of it outside of acknowleding a single issue with some of this though

@camertron

What other issue do you consider relevant for whether LLMs are incompatible with free software?

@azonenberg @conservancy

@wyatt @conservancy @Li Yeah. Abusers are never entitled to being accepted-back, especially not by the party they've abused or other parties who have a duty not to make an unsafe environment for the parties they've abused.

If someone is going to be accepted-back, they need to have done the work to undo or compensate for the harms they caused and demonstrate commitment to change. Not just say "I changed my mind! I'm not bad anymore!"

@troed @azonenberg @conservancy
So you're saying it _is_ a copyright-washing machine? Train it on copyleft code and get CC0 out.

@azonenberg what are your arguments for that opinion?

There is popular Free Software with LLM-generated code, like Linux and systemd. They don't become closed-source or explode, so your statement is not obviously true.

Would it change your opinion when an LLM would be ethically trained and more reliable?

@conservancy

@dalias @conservancy I'm sorry if this came out as a threat. It was not meant as such. What I mean is that I'm confident that I could and that leaves me to believe that a lot of other people might as well. At that point you get to a cat and mouse game where some innocent people might get caught in the crossfire. If there was a surefire way for you to block LLM code I'd support that even. More power to you. What I fear is that we all will just be ever more angry, frustrated and distrustful.

@wyatt @conservancy @Li OTOH if someone is *privately* using "AI" tools for non-codegen purposes, not pushing others to engage with any of that, and being exceedingly cautious to disclose anything that might be unacceptable or tainting to any projects they engage with, I'm not going to shame them.

I'm going to thank them for being decent and respecting our consent even if I deem the tools they're using unethical and harmful.

@troed @azonenberg @conservancy The SFC describes themselves as "a nonprofit organization centered around ethical technology." LLMs are inherently an unethically-made technology. The number of individual ethical concerns is enormous.

@camertron

I don't agree with that blanket claim.

@azonenberg @conservancy

@conservancy

This shit is ass and says nothing. You have no stance and no principles. You are a doormat for any corporation who wants to step on you.

@chris @conservancy What I mean is that the way of thinking is inherently gross. It's the same mindset as "I bet I could slip off the condom without them noticing". If someone is using this line of reasoning, I am going to deem them unsafe and untrustworthy, and it's going to be nearly impossible to undo that perception.

@dalias @conservancy I am sorry but this goes too far. I am honestly shocked by this analogy. Pushing a discussion about ethics in open source in this direction is just tasteless and over the top.

@chris @dalias @conservancy For the past three posts you've essentially been saying nothing but: "I totally could violate your boundaries without repercussions. I could slip in things against your will and you wouldn't even notice. Think about just how easily I could do nonconsensual things if I wanted. Really picture it."

And you *seriously* wonder why it reminds us of that? Really? And then to have the gall to act offended when called out... it's really not a good look.

@FritzAdalis

I'm saying there's a chance all corporate written code using LLMs becomes free software.

@azonenberg @conservancy

@troed that does not mean they have to release it.

Copyleft licenses like GPL forces companies to share their code changes, at least with customers.

Which leads to more freedom for users. The goal of Free Software.

A practical example would be OpenWrt.

@FritzAdalis @azonenberg @conservancy

@chris @dalias @conservancy

So when people say they don't want something, your reaction is to say how you can probably get them to accept it anyway?

Follow-up question: why are you not an asshole? You are openly hypothesizing about tricking people and ignoring their boundaries.

@conservancy calling this a "new era" is highly disingenuous. This era will be extremely short, it'll only last until the funding runs out.

@gabrielesvelto Local open weight models are already good enough. They won't disappear and running them on a regular gaming GPU doesn't cost more than playing a game.

@conservancy

@troed @conservancy who's gonna train them and with what data? Running the models is only part of the problem

@gabrielesvelto They already exist. @conservancy

@troed @conservancy if you're going to use a model for coding then it's better be up-to-date or it will be completely useless

@gabrielesvelto I don't think you know how LLMs work. You don't use them for the knowledge they contain, like a database, but as "thinking engines" that are able to work on data and information you supply.

@conservancy

@troed @conservancy I know exactly how they work and more importantly I've seen how they can be used given an appropriate harness and what are their flaws even when working within those conditions. Some background in case you're curious: https://blog.mozilla.org/en/privacy-security/ai-security-zero-day-vulnerabilities/

@gabrielesvelto If you know exactly how they work you would already know that they're not going anywhere just because the current insane funding will come to an end.

@conservancy

@troed @azonenberg @conservancy There's a legal claim to be made, or at least the New York Times believes there is (they're suing OpenAI, Anthropic, and Microsoft).

But the biggest ethical concern with the technology itself is that the models were all trained with essentially stolen data and constitute large-scale IP theft.

@camertron

All developers who learn how to code do so by looking at code written by others. It doesn't mean anything was stolen.

@azonenberg @conservancy

@davidak

LLMs are a scam run by fascists. they'll never be ethical or reliable. Projects allowing LLM generated patches should be held accountable.

@azonenberg @conservancy

@troed @azonenberg @conservancy Most authors are happy to hear their work was seen by human beings who learned from it - that's probably why they wrote it in the first place.

If you instead tell those authors their work was used by hyperscaler techno-fascists to build a labor-displacing text extruder... they might feel differently.

Like so many modern issues, the problem is consent.

@camertron From the "hyperscaler tehnofascists" I'm guessing you're in group 1 here - correct?

https://swecyb.com/@troed/116725135832925252

If so please go rant at someone else.

@azonenberg @conservancy

@emma i think that's too one-sided, ideological and not helpful in a discussion.

What if it's not?

There are LLMs trained at universities. They could be trained on freely licensed data. But they are tiny and useless.

I think other architectures could be more reliable, like a Large Concept Model.

I think i'm on your side, but i'm also excited for the potential of thinking machines, as a common good.

@azonenberg @conservancy

@troed @azonenberg @conservancy Nope! I'm a software engineer with 20+ years of experience. Here's my GitHub: https://github.com/camertron

I use the tools every single day (as required by my employer), and find them useful.

That doesn't mean I can't also criticize them for the ethical horrors that they are.

Also, your argument that people who don't use LLMs don't know how they work is directly contradicted by the data: https://journals.sagepub.com/doi/10.1177/00222429251314491

@davidak

How about people for their labor instead?

@azonenberg @conservancy

@conservancy

s/volumous/voluminous/ , I think?

Good stance. I appreciate the focus on encouraging newcomers rather than shaming them for the path by which they approached. Still, I think this stance is rather more tolerant of a corrosive set of businesses than is wise; time will tell.

@chris no, it's very apt. It's about consent. If someone says "no, don't do this thing" and you respond with "well but what if I did?" then you have crossed a boundary

Consent is a very large problem in FOSS and has been since its inception.

@dalias @conservancy

@conservancy 💩

@davidak

Sure, I agree that LLM-produced software might not be compatible with copyleft licenses. I don't agree that public domain isn't free software.

@FritzAdalis @azonenberg @conservancy

@troed i did not say that LLM-generated code is not compatible with Copyleft licenses. Linux has LLM-generated code and is GPLv2. That seem to work.

I also did not claim that Public Domain is not Open Source. FSF thinks it is (when code is published) while OSI is not.

https://www.gnu.org/philosophy/categories.html
https://opensource.org/blog/public-domain-is-not-open-source

My point is that it doesn't matter if code is technically public domain, when it is not public!

@FritzAdalis @azonenberg @conservancy

@davidak @troed @FritzAdalis @azonenberg @conservancy

> I also did not claim that Public Domain is not Open Source. FSF thinks it is

I havent read the entire conversation very well, but i'd just like to say that the FSF has little to do with OSS.

@chris @dalias @conservancy definitely overrating your ability to trick people here

@dalias @conservancy I always get „But your licence allows it. Change your licence!“ As if any AI tool cared. As if the right licence existed. As if any normal human being were able to create their own custom-made licence.

@jlink @conservancy No FOSS licenses except 0BSD and similarly minimal ones allow it. The vast majority have requirements to preserve copyright notice & attribution, which fundamentally can't be done when incorporating scraped code into LLMs.

@chris @conservancy I don't think addressing the fact that the root problem with the tech industry is rape culture is "over the top". It's the same mindset and a number of the people behind it (e.g. Sam Altman) are outright literal rapists.

@dalias @chris @conservancy not that it really matters here because I don't think that @chris is arguing in good faith but a less extreme analogy that I've read (and that I can't find the source of) is that if I invited someone in my house and said "these here on the shelf are our family heirlooms, please don't touch them" and they said "oh but how would you ever know if i did" the sensible course of action would be to kick them out

@chris You should be shocked. By what you’ve just learned about your own attitude to consent.
@conservancy @dalias

@dalias @conservancy

even shorter:
My home - my rules.

@dalias @conservancy AFAIK that holds for copyright aspects but not usage of a tool / library. And even with copyright some argue that „training“ is not restricted because fair use.

@jlink @conservancy Regardless of whether a compromised court system in some jurisdiction lets that by, the license does not *allow* it. It's a violation of the authors' consent and is not us "failing to use the right license".

@camertron

Apologies. I'm very tired from a bit too many interactions with anti-AI fanatics and I try to simply avoid engaging with them now.

I agree that there are companies who most definitely trample all over ethical boundaries - but it's not something inherent to the technology itself. When this gigantic bubble bursts (soon, I hope) we'll see more sane practices taking place. Until then I strictly use local models as well as only paying a monthly subscription to Mistral AI, considering them to be the "more ethical" of the bunch.

I've release public code for 40 years now and I truly don't care whether it's a human that sees it or if it's used to train LLMs. The end result is that my work is used to help aid others.

@azonenberg @conservancy

@wrosecrans @conservancy Wasn't it glue?

@emma this is the moderate/compromise position

a strict position would be to also permanently ban individuals based on past usage and to make liberal use of public shaming

@conservancy

@conservancy

Is the recommendation "Fucking stay away from that shit!" ? No?
Well, then kindly fuck off.

@chris @dalias @conservancy I'm absolutely certain that you can create a PR using stolen proprietary code that open source maintainers will not be able to detect. That doesn't mean open source projects should welcome stolen proprietary code.

@chris @dalias @conservancy unfortunately…

I just think it's sad if we started mistrusting each other over LLMs

… we’ve been there for 2+ years already. For code, for writing, for images. And it’s really sad that that has become a necessity, but I fully blame the fashtech bros.

I'd rather assume it's a human

I’d like to be able to do that, but these days, you cannot.

@mathew @chris @conservancy Well, it depends on the project and scope of the change. For projects that are particular about coding style or quality, it's unlikely that anything copied from proprietary code would be anywhere close to acceptable without significant revisions/rewrites after submittion & review.

@chris @dalias
this is just the same dipshit genx playbook where people start rambling about witch hunts and theoreticals and then run straight to "well if I'm going to be under scrutiny for possibly doing something wrong I'm just gonna do it"
@conservancy

@mancube @chris @dalias @conservancy oy, leave GenX out of that

@jlink @dalias @conservancy it actually doesn’t allow that, fun thing. Not that they care.

@jlink @dalias @conservancy the excuse used for “training” is the Text and Data Mining exception to copyright law.

Which demands that an explicit machine-readable opt-out (sad and bad as that is) is honoured, which the fashbros don’t.

Which is for “automatisierte Analyse von einzelnen oder mehreren digitalen oder digitalisierten Werken, um daraus Informationen insbesondere über Muster, Trends und Korrelationen zu gewinnen” and decidedly not “generative AI” (regurgitative). It does not even cover distribution of a model so made.

The second you use a model for regurgitating instead of analysis, the TDM exception you might have been under is null and void.

@dalias @jlink @conservancy yeah, that (consent) too

@wouter @dalias @conservancy you’ve seen how some people (I’m not going to name them here, even if most of the related mailing list discussion is public) react. They’ll never accept that.

@wouter @dalias @conservancy yes, not at first sight = allow for honest mistakes, educate them, help them get into the thing for real if they have an actual interest. Definitely.

But people who knowingly defy that?

@dalias
more addled genx nonsense from the same people who want to tell you it's wrong to close the doors to your community to sex pests, fascists, and other creeps
@conservancy

@mancube @dalias @conservancy oy, leave GenX out of that!

@wouter @conservancy I don't think that's the point, or if it is, they're dangerously unclear about the point in a way that will be weaponized by AI pushers.

"Don't shun" does not mean "don't immediately shun without giving them a chance to correct their behavior". Even if it did, that's a form of "assume good intent", which is *always* an abuse-enabling clause in CoC-like contexts.

@mirabilos @dalias @conservancy Morally I'm fully on your side. I just don't think that eventually highest jurisdictation (anywhere in the Western world) will decide against AI tech.

@conservancy @jlink @dalias doesn’t make it right, or even legal

@conservancy @dalias @jlink (something can be both illegal and not held in by a court at the same time; easiest way is too low value, possibly combined with lower court declining to act and the value is insufficient for higher court, even if recurring, BTST)

@mirabilos @conservancy @jlink And means whoever is using this shit is going to have some fun not being able to operate in the jurisdictions that don't bend to the AI bros. 😈

@dalias @mirabilos @conservancy @jlink

Umm... are there any of those? I even see European countries acting all over the moon over this crap. 🤦‍♂️

@conservancy @jlink @rl_dane @dalias I can imagine that, should these countries figure out a way to extract money from such companies, there suddenly might be some

… but not atm, no. Our current right-wing-barely-not-extremist government is likely to stick in the other fashs’ common cesspool.

@wouter @conservancy It doesn't matter what you or I think it means. It matters how AI pushers will weaponize it.

And in any case, shunning on first sight is well within a community's rights. Nobody is entitled to tell them they have to entertain and try to fix folks who very well might just be bots.

@wouter @conservancy This isn't about what you or I would personally rather do. It's about what is being pushed on projects as a norm for how they should handle a situation. And this is not something that should be pushed upon them. Whether they want to educate these people or immediately click the block button is entirely their business not ours or the Conservancy's business to tell them what to do.

@troed @azonenberg @conservancy No need to apologize, there's a lot of nuance missing from the larger gen AI conversation, so I can't blame you for feeling defensive 😔

While I agree that much of the ethical blame lies with eg. individual frontier model companies in this current hype cycle, technology does not exist in a vacuum and never has. All technological progress comes at a cost, and the people who bear the brunt of that cost are seldom the ones who benefit.