👀 … https://sfconservancy.org/blog/2026/apr/15/eternal-november-generative-ai-llm/ …my colleague Denver Gingerich writes: newcomers' extensive reliance on LLM-backed generative AI is comparable to the Eternal September onslaught to USENET in 1993.
-
(2/5) … In https://sfconservancy.org/blog/2026/apr/15/eternal-november-generative-ai-llm/ ,
Denver's key points are: we *have* to (a) be open to *listening* to people who want to contribute #FOSS with #LLM-backed generative #AI systems, & (b) work collaboratively on a *plan* of how we can solve the current crisis.Nothing ever got done politically that was good when both sides become more entrenched, refuse to even concede the other side has some valid points, & each say the other is the Enemy. …
@bkuhn For those who are acting in good faith, and willing to contribute in healthy ways - yes, it's absolutely worth while to talk to them, and try to get them to contribute in good ways. If you can, have time, are not drowning in so much slop that you can't tell who means well and who just needs to be blocked/banned, etc. - integrating people into the community is a lot of work, and a lot of people maintaining and making free software are already doing a lot of work for free as it is. (1/?)
-
@kees @josh @silverwizard @ossguy @bkuhn @karen @wwahammy @xgranade
I suspect (as some scientists do as well) that this is not a cultural, but a neurological phenomenon. So I think this is really on a totally different level.
The skill erosion I am talking about has nothing to do AT ALL with critical thinking.
Best case it's just that if of two synaptic circuits (use the translation tool vs retrieve from memory) the one which wanted to activate, but then got not chosen, is actively deleted or weakened. My understanding is that this is how biological brains work / learn.
The worse alternative is that the output of LLMs has some yet not sufficiently described hidden quality which poisons neuronal networks that process them.
One hint in this direction is, that LLM models, who consume the production of other LLM models in training, collapse. That's a clear indicator that on some level LLM output is observably different from human language production, though we as humans have on average a hard time telling the difference.
And this is what I am talking about: Not a loss of cultural techniques or of learned skill by atrophy or not being taught anymore, but the poisoning of neuronal networks by input they cannot firewall because they have not evolved to recognize it as a hazard.
@glitzersachen @josh @silverwizard @ossguy @bkuhn @karen @wwahammy @xgranade
Right, yeah, this is why I've cautioned people about *how* they use LLMs. You've distilled it more clearly and lines up with my own intuition that reminds me about how human memory systems work: retrieval is effectively erasure, so "remembering" requires retrieval and storage. Research into treating PTSD (IIRC?) and such found that blocking storage (with drugs or EM) and then triggering recall would wipe memories. You're describing a potentially purely experiential way to do this, which is terrifying.
I feel like using an LLM can lead to a Dunning-Kruger like effect, in that you think you know what it did, but you don't. And this belief satisfies the need/instinct to learn/know what happened without having actually done so. (Reminds me of making a TODO list and now the Dopamine hit from that kills the need to actually *do* the list.)
-
@bkuhn For those who are acting in good faith, and willing to contribute in healthy ways - yes, it's absolutely worth while to talk to them, and try to get them to contribute in good ways. If you can, have time, are not drowning in so much slop that you can't tell who means well and who just needs to be blocked/banned, etc. - integrating people into the community is a lot of work, and a lot of people maintaining and making free software are already doing a lot of work for free as it is. (1/?)
@bkuhn That does not mean that LLM-generated code, assets, or outputs can be allowed in free-software projects. Which is an important distinction.
If someone wants to contribute, that's great - point them to resources of how to do so, how to make submissions that can be accepted. If they won't contribute without using claude or whatever, then their contributions must be refused.
The ethical, environmental, public health, freedom/human rights, issues of LLMs as they exist are too severe (2/?) -
@bkuhn That does not mean that LLM-generated code, assets, or outputs can be allowed in free-software projects. Which is an important distinction.
If someone wants to contribute, that's great - point them to resources of how to do so, how to make submissions that can be accepted. If they won't contribute without using claude or whatever, then their contributions must be refused.
The ethical, environmental, public health, freedom/human rights, issues of LLMs as they exist are too severe (2/?)@bkuhn to be acceptable in free software. I would go so far as say the free software definition should be ammended, to exclude any component generated by an LLM or similar generative program that is not:
- Purely deterministic OR
- Completely free, including all training data, weights, source code, training processes, etc. such that a user (with sufficient resources) could recreate the process from the ground up, in the same way that a user can re-compile GCC, & uses only (3/?) -
@bkuhn to be acceptable in free software. I would go so far as say the free software definition should be ammended, to exclude any component generated by an LLM or similar generative program that is not:
- Purely deterministic OR
- Completely free, including all training data, weights, source code, training processes, etc. such that a user (with sufficient resources) could recreate the process from the ground up, in the same way that a user can re-compile GCC, & uses only (3/?)@bkuhn Ethically-acquired (so, no DoSing an artists website for it, no ignoring a robots.txt to scrape it, etc) training data.
This doesn't solve the *other* ethical problems, but, the nature of these generators is that without freedom in the full chain from data to output as a minimum, they should be excluded from the free software definition - in the same way inserting a binary blob into the linux kernel makes it at least partially non-free. The LLM is a non-free black-box without those(4/?)
-
@bkuhn Ethically-acquired (so, no DoSing an artists website for it, no ignoring a robots.txt to scrape it, etc) training data.
This doesn't solve the *other* ethical problems, but, the nature of these generators is that without freedom in the full chain from data to output as a minimum, they should be excluded from the free software definition - in the same way inserting a binary blob into the linux kernel makes it at least partially non-free. The LLM is a non-free black-box without those(4/?)
@bkuhn parts being as free as the source code, and including their outputs should be treated as a binary blob - since there is no way to investigate the process behind how it ended up there or was created. It can't even be meaningfully reverse-engineered in most cases.
I'd also add to any project refusal of any generated content that carries the climate, freedom, rights, labour rights, etc. concerns as well - but at a minimum, the outputs of an LLM can not be considered free. (5/5)
-
@bkuhn parts being as free as the source code, and including their outputs should be treated as a binary blob - since there is no way to investigate the process behind how it ended up there or was created. It can't even be meaningfully reverse-engineered in most cases.
I'd also add to any project refusal of any generated content that carries the climate, freedom, rights, labour rights, etc. concerns as well - but at a minimum, the outputs of an LLM can not be considered free. (5/5)
@bkuhn The *people* coming in, if they mean well and will contribute in ethical ways, are fine, and worth welcoming in.
But the LLMs themselves - the use of the tool - as it exists undermines the ethical core of the free software movement, and carries too many other ethical problems to be acceptable.
(6/5) -
@bkuhn The *people* coming in, if they mean well and will contribute in ethical ways, are fine, and worth welcoming in.
But the LLMs themselves - the use of the tool - as it exists undermines the ethical core of the free software movement, and carries too many other ethical problems to be acceptable.
(6/5)@bkuhn This leaves room for an ethical, actually free, version of the tech, should it appear at some point, which is a compromise, my instinct is that there can be no ethical version, but, I could be wrong.
As-is though, the LLMs & Genrative systems in use are a black box - even 'open source' ones, if they do not also provide full access to the training data & methodology, including them in free software is no better than proprietary code. The definitions & licenses need to reflect this. (7/5) -
@glitzersachen @josh @silverwizard @ossguy @bkuhn @karen @wwahammy @xgranade
Right, yeah, this is why I've cautioned people about *how* they use LLMs. You've distilled it more clearly and lines up with my own intuition that reminds me about how human memory systems work: retrieval is effectively erasure, so "remembering" requires retrieval and storage. Research into treating PTSD (IIRC?) and such found that blocking storage (with drugs or EM) and then triggering recall would wipe memories. You're describing a potentially purely experiential way to do this, which is terrifying.
I feel like using an LLM can lead to a Dunning-Kruger like effect, in that you think you know what it did, but you don't. And this belief satisfies the need/instinct to learn/know what happened without having actually done so. (Reminds me of making a TODO list and now the Dopamine hit from that kills the need to actually *do* the list.)
@glitzersachen @josh @silverwizard @ossguy @bkuhn @karen @wwahammy @xgranade
I lump my experiences of software engineering use of LLMs into 3 modes:
1) "work together", I am watching everything it is doing, reviewing every step, and contributing to the result in tandem. This doesn't feel to me like anything is being eroded on my end. But I'm also a deep sceptic of its output.
2) "do the thing I know how to do for me", this is super dangerous, as I think I'm solving problems I am familiar with, but I didn't follow the results closely and I'm left with deep erosion of my comprehension of both problem and solution.
3) "vibe coding", I have no idea what it is doing with a thing I don't know about and I know I have no idea what it is doing. This doesn't seem to erode anything. It does create a new problem for me, though, if the LLM can't solve some problem because also neither can I.
I've felt #2 a few times, and I had the alarm bells in place to shift myself back to #1, which required doing full review and looking back through the reasoning and checking the work. The risk of being drawn into #2 is high given the sychophancy of the models, but I think my suspicion of it has helped avoid this a bit.
(And perhaps I am more deluded than I think.)#3 I have done for educational/amusement purposes, but it's an uncommon mode for me because what's the point of creating a thing I don't understand and can't fix?
("I can quit any time!")
-
@glitzersachen @josh @silverwizard @ossguy @bkuhn @karen @wwahammy @xgranade
Right, yeah, this is why I've cautioned people about *how* they use LLMs. You've distilled it more clearly and lines up with my own intuition that reminds me about how human memory systems work: retrieval is effectively erasure, so "remembering" requires retrieval and storage. Research into treating PTSD (IIRC?) and such found that blocking storage (with drugs or EM) and then triggering recall would wipe memories. You're describing a potentially purely experiential way to do this, which is terrifying.
I feel like using an LLM can lead to a Dunning-Kruger like effect, in that you think you know what it did, but you don't. And this belief satisfies the need/instinct to learn/know what happened without having actually done so. (Reminds me of making a TODO list and now the Dopamine hit from that kills the need to actually *do* the list.)
-
People shouldn't be *abused*, ever. If people are *shilling* AI and trying to force it on others, they might deserve some amount of shame and disapprobation. But nobody deserves abuse.
@josh @silverwizard @ossguy @bkuhn @karen @wwahammy
Agreed. I only want to make clear that it should not be clear flying for AI shills. A certain amount of actually painful headwind should meet them.
On the other side I consider what some middle management does to their employees with AI abuse (not sure about the apllicability of the English word here. I am not a native speaker).
And in line with Popper's paradox of tolerance I wonder whether abuser _in a position of power_ should really be countered with slight slap on the hand and mild mannered words.
After all it was this kind of behavior in such a power constellation that triggered revolutions historically. And I have a hard time rooting for the oppressors.
-
@davidgerard @wwahammy @silverwizard @firefly_lightning @cwebber Yes, which is why it's important to allow people to identify when they have used LLM/AI assistants to help. New contributors will see this is the norm, and then it will be easier to help them, because we'll know a bit about where any potential knowledge gaps might be coming from.
If we "ban" LLM/AI-assisted contributions, people will use them anyway but hide their use, which is a trickier problem to solve.
@ossguy @firefly_lightning @davidgerard @wwahammy @cwebber so is the theory "we should accept people who have been abused by big tech and think LLM usage is ok, and we must help them understand why it's not", or is "we must accept LLM code commits because there's people who think they're ok and the longterm damage is a risk we'll just need to figure out later"?
I keep thinking it's the very reasonable first one, which I don't think is up for debate. But you and Bradley keep implying it's the second, which *is* up for debate.
-
@cwebber I think maybe you missed https://sfconservancy.org/blog/2026/mar/04/scotus-deny-cert-dc-circuit-thaler-appeal-llm-ai/ where #SFC analyzed that situation?
Also, follow @ai_cases & see the *firehose* of litigation on this & remember the “Work Based on the Program” issue under GPLv2 has still never been litigated directly but lots of cases about 100% proprietary software have bolstered GPL's strength.Big Content has legal battles with Big Tech on 100s of fronts rn. Yes, we're adrift on their sea, but the situation is not as dire as you imagine.
@bkuhn @ossguy @richardfontana Continuing here, because it's the relevant subthread.
I am sympathetic to choosing to narrow a topic. However, the post, in implying that we should start accepting partially AIgen contributions, inherently pulls in the topic of whether or not that is legally safe.
Yes, I have read the previous Conservancy post about the existing cases. This partly contributes to my surprise and confusion about the post.
Acknowledging that the plan is to have continued conversations and meetings about this, I still feel it is important to lay down my current concerns, even before such a meeting. I am leaving the "quality of contributions" and many other details out of here, and instead focusing on whether of not it is *safe to accept* contributions on copyright grounds at the moment, and what the implications of thinking on that are.
(cotd)
-
@bkuhn @ossguy @richardfontana Continuing here, because it's the relevant subthread.
I am sympathetic to choosing to narrow a topic. However, the post, in implying that we should start accepting partially AIgen contributions, inherently pulls in the topic of whether or not that is legally safe.
Yes, I have read the previous Conservancy post about the existing cases. This partly contributes to my surprise and confusion about the post.
Acknowledging that the plan is to have continued conversations and meetings about this, I still feel it is important to lay down my current concerns, even before such a meeting. I am leaving the "quality of contributions" and many other details out of here, and instead focusing on whether of not it is *safe to accept* contributions on copyright grounds at the moment, and what the implications of thinking on that are.
(cotd)
@bkuhn @ossguy @richardfontana So the question is: is it safe, from a legal perspective, given the current state of uncertainty of copyright of such contributions, to encourage accepting such contributions into repositories?
Now clearly, many projects are: the Linux kernel most famously is, and their recent policy document says effectively, "You can contribute AI generated code, but the onus is on you whether or not you legally could have".
Which is not very helpful of a handwave, I would say, since few contributors are equipped to assess such a thing. I've left myself and three others addressed in this portion of the thread, and all of us *have* done licensing work, and my suspicion is, *especially* based on what's been written, that none of us could confidently project where things are going to go.
-
@bkuhn @ossguy @richardfontana So the question is: is it safe, from a legal perspective, given the current state of uncertainty of copyright of such contributions, to encourage accepting such contributions into repositories?
Now clearly, many projects are: the Linux kernel most famously is, and their recent policy document says effectively, "You can contribute AI generated code, but the onus is on you whether or not you legally could have".
Which is not very helpful of a handwave, I would say, since few contributors are equipped to assess such a thing. I've left myself and three others addressed in this portion of the thread, and all of us *have* done licensing work, and my suspicion is, *especially* based on what's been written, that none of us could confidently project where things are going to go.
@bkuhn @ossguy @richardfontana Part of the problem here is that the AI companies have set the stage themselves. Their presumption is that it's fine to absorb effectively all open and "indie" content, and that this is entirely fair to pull into a model without any legal implications, whereas potentially yes, you may need to "license" something that looks like a Disney character. In the land of code, I also sense that Microsoft is perfectly fine with the idea that you can "copyright launder" a codebase from the GPL to perhaps the public domain, but if someone did that to their own leaked source code, they would be very upset.
Meanwhile, a friend of mine who works in films has said that he keeps hearing rumors that OpenAI would like a cut of stuff made with their stuff. We should presume tthat true.
Regardless, I'm sure everyone on this thread wants an *equitable* situation for proprietary and FOSS licensing. I'll expand on that more in a moment though.
-
@bkuhn @ossguy @richardfontana Part of the problem here is that the AI companies have set the stage themselves. Their presumption is that it's fine to absorb effectively all open and "indie" content, and that this is entirely fair to pull into a model without any legal implications, whereas potentially yes, you may need to "license" something that looks like a Disney character. In the land of code, I also sense that Microsoft is perfectly fine with the idea that you can "copyright launder" a codebase from the GPL to perhaps the public domain, but if someone did that to their own leaked source code, they would be very upset.
Meanwhile, a friend of mine who works in films has said that he keeps hearing rumors that OpenAI would like a cut of stuff made with their stuff. We should presume tthat true.
Regardless, I'm sure everyone on this thread wants an *equitable* situation for proprietary and FOSS licensing. I'll expand on that more in a moment though.
However, it's not actually the laundering angle I am concerned with here entirely, it's whether we're turning FOSS codebases into potential legal toxic waste dumps that we will have a hell of a time cleaning up later.
The previous Conservancy post, which @bkuhn linked upthread, indicates that Conservancy does indeed consider the matter unsettled.
Current LLMs wouldn't "default to copyleft", since they also include all-rights-reserved mixed in there. If the result of output of these systems is a slurry of inputs which carry their licensing somehow, their default licensing output situation is one of a hazard.
I note that @bkuhn and @ossguy seem to be hinting at hoping a "copyleft based LLM" with all-copyleft output it a winning scenario. I'm going to state plainly: I believe that's an impossible outcome.
-
However, it's not actually the laundering angle I am concerned with here entirely, it's whether we're turning FOSS codebases into potential legal toxic waste dumps that we will have a hell of a time cleaning up later.
The previous Conservancy post, which @bkuhn linked upthread, indicates that Conservancy does indeed consider the matter unsettled.
Current LLMs wouldn't "default to copyleft", since they also include all-rights-reserved mixed in there. If the result of output of these systems is a slurry of inputs which carry their licensing somehow, their default licensing output situation is one of a hazard.
I note that @bkuhn and @ossguy seem to be hinting at hoping a "copyleft based LLM" with all-copyleft output it a winning scenario. I'm going to state plainly: I believe that's an impossible outcome.
@bkuhn @ossguy @richardfontana Rather than focus on the GPL, let's choose a different copyleft license. In fact, let's choose a gradient of licenses.
- CC0's public domain declaration w/ minimal fallback license
- CC BY
- CC BY-SAImagine for a moment an LLM trained entirely on the above three licenses, and then one that's CC BY and CC0, and then one that's just CC0.
Let's look at both extremes and then we'll find out the real dangers come from observing the middle.
-
@bkuhn @ossguy @richardfontana Rather than focus on the GPL, let's choose a different copyleft license. In fact, let's choose a gradient of licenses.
- CC0's public domain declaration w/ minimal fallback license
- CC BY
- CC BY-SAImagine for a moment an LLM trained entirely on the above three licenses, and then one that's CC BY and CC0, and then one that's just CC0.
Let's look at both extremes and then we'll find out the real dangers come from observing the middle.
@bkuhn @ossguy @richardfontana First let's imagine the only-CC0 based LLM.
I would fully agree that no matter the law and legal case law passed and established, the CC0 based input LLM is clearly effectively in the public domain, or like CC0 itself, equivalent to it. This one is relatively simple.
Let's make things more complicated.
-
@bkuhn @ossguy @richardfontana First let's imagine the only-CC0 based LLM.
I would fully agree that no matter the law and legal case law passed and established, the CC0 based input LLM is clearly effectively in the public domain, or like CC0 itself, equivalent to it. This one is relatively simple.
Let's make things more complicated.
@bkuhn @ossguy @richardfontana Regarding the one containing CC0, CC BY, and CC BY-SA, the situation is more uncertain and seems highly affected by legal outcomes in upcoming law and cases to be set. There is the possibility that indeed, the LLM is considered a slurry of inputs and this is legally acceptable, and effectively any output which is not verbatim of its inputs in some way is effectively under the public domain.
Now, of course, the problem is that we don't have to just worry about the US, we have to worry *internationally*. When considered from this angle, that FOSS is an international endeavour, this hope that things are in the public domain feels a lot dicier.
The assumption is that then this effectively leads to the output being under the terms of CC BY-SA. This is fine, great even, right?! Because effectively everything is share-alike (Bradley I don't wanna get into whether BY-SA is copyleft or something weaker). We slap CC BY-SA on the output, it's fine. Right??????
-
@bkuhn @ossguy @richardfontana Regarding the one containing CC0, CC BY, and CC BY-SA, the situation is more uncertain and seems highly affected by legal outcomes in upcoming law and cases to be set. There is the possibility that indeed, the LLM is considered a slurry of inputs and this is legally acceptable, and effectively any output which is not verbatim of its inputs in some way is effectively under the public domain.
Now, of course, the problem is that we don't have to just worry about the US, we have to worry *internationally*. When considered from this angle, that FOSS is an international endeavour, this hope that things are in the public domain feels a lot dicier.
The assumption is that then this effectively leads to the output being under the terms of CC BY-SA. This is fine, great even, right?! Because effectively everything is share-alike (Bradley I don't wanna get into whether BY-SA is copyleft or something weaker). We slap CC BY-SA on the output, it's fine. Right??????
@bkuhn @ossguy @richardfontana Except, I actually believe this scenario isn't legally viable. And it's easier to understand if we scale back to the middle case.
Let's now look at the LLM trained on CC0 and CC BY. Because it's the BY aspect that makes everything complicated.
There is *NO WAY* in current LLM technology, nor I believe from studying how neural networks work, any viable computationally performant LLM, that they can track provenance. The BY clause cannot be upheld.
This isn't a theoretical concern for me; someone built another vibecoded Scheme-to-WASM-GC compiler that looks an awful lot like Spritely's own Hoot compiler in places. They didn't attribute us. They probably didn't know. But like many FOSS licenses, Apache v2 does require certain levels of attribution to be upheld. Most FOSS projects do.
You can't uphold the CC BY requirement, as far as I can tell.
