That’s why many program EULAs have an anti-reverse-engineering clause. (Not that these clauses are necessarily enforceable).
1 Like
If that’s true, then it’s going too far for my taste of satire as well 
Is it true though? I really didn’t dig into this and only heard about it from this post. All I have seen is that website that claims that they offer a service.
1 Like
I believe this is the same website that ThePrimeagan made a video about.
It seems satirical, but is not. He even tested it by paying for it to implement some small Java/Type-Script package, and it churned out a working result that passed the same test suite.
I have serious doubt on how well it would scale with a non-trivial project, but the fact remains they are actually selling the service for money, and it is doing what it is advertised to do.
5 Likes
Crazy! I’ll watch that video. Thanks for pointing it out 
1 Like
I’ve been trying to think and feel like this, but the problem I’m running into is I love building software for people to use, I love creating things for other software crafts people, and I love working creatively with other people.
I hate AI, a lot. So much that I recently quit my job of 5 years as CTO of a VC-backed (
) tech company over it. Spending my days prompting some matrix multiplication in a trenchcoat to perform my chosen craft is not the career I signed up for. Or more accurately, managing other engineers spending their days writing prompts.
I thought I would finally have the time to dedicate to building CommanderDB, the database I have been dreaming about creating for a few years. Instead, I’m finding AI totally demoralizing. Not because I believe it can produce a genuine substitute for well crafted software, it can’t. No, it’s because people don’t seem to care about software, about quality. Constantly seeing ads for Base44 on YouTube reminds me anything I create will be a single mote in a dust storm of terrible apps. Reading source code for new packages I come across and realizing it’s AI generated makes me question if any humans will ever use any packages I create, or will my audience be entirely “agentic“…
The answer can’t be “AI will make people care about quality and craftsmanship again“ because although the outputs of AI aren’t a genuine substitute, they flood the market so cheaply and quickly that they erode people’s capacity to discern between genuine human endeavour and AI slop. People will have to give up by either turning their backs on things they love entirely, or giving up the practice of discernment and accepting slop into their lives. We’re already seeing the beginning of this in OSS with the rise of slop PRs.
I really hope there’s something I’m missing, because at the moment it seems like I’m mourning not only a career I liked, but also a craft I loved
12 Likes
To the original point, I’ve started hosting my own Forgejo instance on my home server. At the moment, I do have it publicly addressable, although I could take it down at any time and keep it private. There’s nothing particularly exciting in my private instance at the moment.
I’ve stopped using GitHub entirely for anything new, and I will only use public services like Codeberg for things I want to share, but yeah, even sharing what I’ve created I’m finding I feel similarly – I only want to share it with people, not these plagiarism-powered Dunning-Kruger amplifiers.
5 Likes
I think you struck gold here, and its absolutely Plato’s cave all over again xd. There is a better world - with not just better software but better everything - but most of people are happy to be chained and watching shadows of what that reality could be.
IMO, people are tasteless in several aspects of file, and its a combination of not having enough criteria and not knowing it exists a better world out there. I know that food comparisons are very overused but I’ll go with that regardless: my mum always says that people will eat absolute dogshit and not realize it, and it’s really not a matter of taste, but a matter of perception. If you never tasted real food made with love, you cannot realize you are missing that and that you prefer it. And even if you know that exists, others telling you this food you eating is actually the best is going to be absolutely fine with you is able to gaslight you more often that not.
Going back to software, what made my mind click was watching Casey Muratori’s talks about OOP and the continuous degrading software quality. Until that point, I was incapable to realize how much modern software sucks and how lower our society bar has actually been lowered. Until that does not get adressed, good luck making people care about not eating shit.
5 Likes
It could be worse. At least there are glimpses outside of the cave. So we are provoked, even when hope seems slight and vanishing at times.
Ok, but other than LLVM, Rust, Go, Swift: what have the Romans done for us?
I’m not here to stan for enormous megacorporations, both out of a lack of inclination, and in view of the fact that they neither need me to nor pay me to do so. Just thought it was funny.
This is an emotionally compatible sentence, but not a logically compatible one. Both open source and free software are crystal clear: discrimination against fields of endeavor is not compatible with either framing of openness / freedom.
We shouldn’t give with clenched hands. It’s bad for the soul.
I have seen, over the 21st century, a growing misapprehension that FOSS is in some way, in any way, an anticorporate movement. It is not, it is not designed to be, and cannot be, due to its core values.
“But copyleft” first off, I don’t use it, second I will address it, hold on.
Amen!
Bizarre that they write so much of it then. Someone should tell them
Also correct. Although I would put it slightly differently: if you want to discriminate against fields of endeavor, you are not writing open source software and should find another hobby.
Swapping “bot” for “actor”, for lack of a better suitably-general word, this is the definition, rather than the spirit, of clean room or “white box” engineering.
Clean room engineering is great! Until two weeks ago, this was highly respected as a way to take the functionality of proprietary software and liberate it.
What changed? We know what changed. It just got much easier to do.
I’d like y’all to at least consider the possibility that the medium-term result of this, is hackers flooding the zone with so much free software that the floor drops to zero on all of it. Free “DefinitelyNotAutoCAD” and so on.
In fact FreeCAD just dropped a suspiciously nice update for 1.1. I mean suspicious! I suspect robots honestly. Think they might be bottin’ over there
Anyway, time for
Copyleft
Because I don’t understand the psychology of getting pressed about bots training on permissive code well enough to address it. That’s just weird. Ask ChatGPT to show you the license for curl without using the Internet, see what happens. It’s in there!
I will be quoting the Gospel According to GNU, 3.0. From the preamble:
Developers that use the GNU GPL protect your rights with two steps: (1) assert copyright on the software, and (2) offer you this License giving you legal permission to copy, distribute and/or modify it.
You know what’s not mentioned here? Reading it and studying it. You know why? Because by common law, and the Berne Convention, that right is granted by the act of publication, applying automatically to any text which has been legally obtained.
To “modify” a work means to copy from or adapt all or part of the work in a fashion requiring copyright permission, other than the making of an exact copy
This is the crux of the matter. Is training a neural network copying? An exact copy is made, but that’s covered. Is it “adapting”?
Let’s put a pin in that one. Say maybe, maybe not. This is an accurate reflection of current legal thinking, so far as I understand it.
You may make, run and propagate covered works that you do not convey, without conditions so long as your license otherwise remains in force.
Suppose that slurping GPL code up into a big ol’ neural net is “adapting” it. Is the adaptation “conveyed”?
For most, no. OpenAI does not convey their models. Meta does! But under a license which is, I’m just going to say, “somewhat ideologically compatible” with GPL. Possible that’s a fruitful avenue of legal action.
You may convey a work based on the Program, or the modifications to produce it from the Program, in the form of source code under the terms of section 4, provided that you also meet all of these conditions
…oops. This sounds like trouble.
And it is. I don’t know if I fooled ya, but this is obviously not happening and I hope FSF sues each of them for all they’re worth.
Dead easy too, you know that stunt Anthropic did where they made Claude “write a C compiler in Rust”? Well they did that by cribbing off the GCC tests as a source of truth.
What do you think would happen if someone said “yo Claude! Here are your ‘acceptance criteria’ write me a C/C++ compiler in C++” then describe the GCC architecture as the way to do it?
Dollar will get you a donut it emits large portions of GCC itself, verbatim.
Can’t stop the signal, Mal
Before you get too excited, I want you to understand where the problem is here. It’s not training on GPL code, that is, and should be, perfectly legal. Not sorry about it because I care about my damn rights, and that includes fair use and transformative works.
Rights for me but not for thee? Not into it. Don’t respect it. Oh but they’re a Corporation and a Large one—didn’t ask don’t care. Fields of endeavor. Take it or leave it.
The problem is that they’ll emit recognizable chunks of it on demand, and I have no doubt in my mind that they will, in fact, do that. You can get a chatbot to recite Harry Potter, so you can get it to recite readline.
How they fix that is up to them. They do need to fix it. Maybe it’ll mean that copyleft code will be even more marginalized than it steadily has been over time.
Good. I’ll say it again good. Stop giving with clenched hands. I understand it, and I respect it, but I do not sympathize with it.
Or maybe they’ll figure out how to fix the ‘emit’ problem. Or maybe they’ll ask what copyright you’re working under and emit on that basis. Who knows.
Maybe a bunch of hackers will obviate the whole thing, by setting up a two-stage clean room engineering bot pipeline to turn GPL code into something equivalent and verifiably re-engineered, complete with audit trail, then release it MIT.
Of all the consequences of generative models to be concerned about, and there are so many of those, I truly believe that “what if they train on my code??” is misplaced angst.
I have long preferred permissive licenses, because I consider them generous and optimistic. From where I sit, that’s worked out. Linux is good because it’s Linux, not because of the license: if it weren’t for some ill-timed lawsuits, Linus would probably not have started his own kernel, since 386BSD would have been available, as its descendants are now. If you think the prevalence of Linuces over BSDs is due to licensing, you are historically misinformed, it’s much more particular and path-dependent than that.
Copyleft is pessimism, and it has its place. My go-to example is FDM slicers. If PrusaSlicer wasn’t AGPL, would BambuLab have just stolen it? Of course, they’re Communist bad actors, but they couldn’t, so BambuSlicer is also AGPL, as is my preferred fork of it, Orca. That’s working as intended.
It seems perfectly obvious to me that any technology making it dramatically easier to write software, weakens the case for pessimism. It might wipe out our industry (not bloody likely, see Jevon’s Paradox for why), it might turn Skynet and eat our faces, but it’s not going to affect the human psychological inclination not to pay for that which one can acquire for free.
Sure, FSF should sue the pants of every single one of the big guys. It won’t change what’s coming, not one iota, but it’s the right thing to do.
Anyway. I won’t make this four times longer by trying to grapple with my own opinion of this thing which is sweeping the world, tempting though that is. But I’ll leave you with a palate cleanser.
Here’s Dr. Knuth hanging out with Claude. If you don’t find that moving, don’t tell me so. Imagine living the life Donald Knuth has lived, and getting to see this day! Bet he gave some thought to his old friend John McCarthy, who did not. It’s a blessing, truly.
Also this:
6 Likes
Wow, that is so inspiring to see. I didn’t even know he was still alive and working. And it demonstrates perfectly how to look positively on the problem being discussed. Thank you for all your thoughts!
1 Like
It really does feel demoralizing, doesn’t it? I wish you hadn’t been greeted with a wall of text saying “NO!” Obviously you know that our existing licenses are not compatible with your wish, but we should keep looking for a pattern that is since there are very many of us wishing for the same thing.
1 Like
very well said, I agree with most of what you’ve said.
I was not stating that foss is anti-corporate, I was stating that corporate trend towards anti-foss. That is also a generalisation, not every private corperation will follow that trend.
Many do see how they depend on OSS, and do take advantage of it:
They still benefit from that! “Unpaid labour” is a cheap argument, since volunteers are free to do just that, but the point of them not needing to pay every developer due to it being OSS is still true.
The bigger benefit is by giving us the tools they need us to use in their employment, we train ourselves at no cost to them. They may of course provide training to get to their standard, but it’s still cheaper than training from scratch.
None of that is inherently bad of course.
My clenched hands are to prevent the hoarding of innovations behind closed doors for the profit of the few! Because that is antithetical to the spirit of FOSS that I stated earlier, and which you agreed with.
I misread that as “If PrusaSlicer was AGPL” and was mightily confused for a while.
But it did cause me to go down a rabbit hole of FOSS and china, which was quite pleasant
1 Like
You have my perfect assent, and I assert that my post was fundamentally an emotional expression.
A few days ago, fwiw, I did opt for MIT (closing the loop on my original post), and I’m moving on with life… I don’t even need therapy. Venting in a post was sufficient outlet. 
5 Likes
I like the arguments you’re making here, and I want to believe them, but there’s a few things stopping me.
Both open source and free software are crystal clear: discrimination against fields of endeavor is not compatible with either framing of openness / freedom.
First of all, is it really endeavor? What if we say “discrimination against fields of human endeavour is not compatible with openness“? When an LLM is trained on my code, or Adam Conover’s scripts, it is not done for any specific purpose. Its purpose is latent, and this latent utility is then sold as a service to someone putting it to their purposes. At this point, very little human endeavor is actually occurring. The effort required for a company to sell my human endeavors as their own service has changed by orders of magnitude – it’s a new paradigm and that means we need to reconsider old questions we thought we had answered.
You know what’s not mentioned here? Reading it and studying it. You know why? Because by common law, and the Berne Convention, that right is granted by the act of publication, applying automatically to any text which has been legally obtained.
Again, I question if that’s what LLMs are doing. Our tendency to anthropomorphise everything means that we chose “training“ as the word describing a process of creating a set of weights statistically likely to output text humans perceive as relevant. LLMs are not reading things and studying them. They are not experiencing them, they are not integrating them into a corpus of experiential knowledge, they’re not there late at night, exhausted, reminding themselves why this is important to them and helping the people in their lives.
I even used the pronoun “they“ without thinking to describe a few billion (trillion?) numbers used to predict text. It’s one hell of a loophole in our brains that these things exploit.
Of all the consequences of generative models to be concerned about, and there are so many of those, I truly believe that “what if they train on my code??” is misplaced angst.
Oh, you have misunderstood my anger. I want these things to drown in their own output and collapse. I want them “dead”. I know it’s unrealistic, but I don’t care. These things are rendering human endeavor, creativity, and creation as commodities. For-profit companies supported open source software because it commoditized their compliments, now they have figured out how to commoditize creativity and human endeavour. It doesn’t matter if doing so will extinguish human ingenuity, if it will halt all human progress, creativity, and innovation – all that matters to these companies is that they are the “category kings“ of the collapse of society. They have a fiduciary duty to shareholders, after all.
4 Likes
I forgot to address this point– I don’t think Jevons paradox applies here. We’re not rendering more efficient utilization of a fungible resource. Again, it’s not a genuine substitute, it’s a forgery taking the place of the genuine article. It’s making the genuine article harder to find and harder to use. It destroys the incentives for creating the genuine article in the first place. That’s partly why these things, if left unchecked, will cause what I call “Total Skill Collapse“ – over decades, LLMs destroy the economics that created the experts in a particular field. People stop entering that field, eventually no one in the workforce really possesses those skills anymore. This subsequently starves LLMs of training data, and people capable of discerning if the LLM output is good or bad. Eventually, we have no humans capable of the skill and no LLM. Humanity as a whole loses an entire skillset developed over centuries and destroyed in decades.
6 Likes
It is a generalization and it’s not the sort of point I would care to dispute. I know raising a misconception while replying to you (amongst others) creates an implication that I think you partake in it. Not my intention.
It’s quite a dichotomy, I agree. Large corporations enrich themselves greatly through the use of free software, and let’s be real, the ‘beer’ interpretation is always there, innit.
They’re also major purveyors of it. Corporations pay for the creation and maintenance of a great deal of FOSS software, that’s a fact as well. I hope, at least, that we can continue to strive for a positive sum in that balance sheet, without compromising the values which got us this far.
Well, and I understand that too. Part of staying within the coalition is that, sometimes begrudgingly, I agree that the limitations, which the reciprocity clause in copyleft imposes, can be construed as in the service of freedom. I think Stallman’s point of view on this kind of thing is upside-down, but that’s just me. I have pretty good theory of mind and can understand why others see it otherwise.
The crux of it is whether your first sentence, which is generally-speaking a thing which does happen, and is bad, is in particular a good description of the Whole LLM Thing. Without implying that in writing that sentence you meant it to be.
Yes.
Training a neural network is a human endeavor. I don’t think this point can be successfully disputed. Them GPUs do not manufacture, purchase, installl, or run themselves.
This is frankly insulting to the work which AI engineers do. I don’t need to take offense on their behalf, but I don’t think you thought this through.
LLMs don’t do. They’re computer programs.
Which you get, this a pure interstitial because that would be an unfair place to cut you off:
The tendency (which is to a significant degree cynical, on the part of Big AI) to choose loaded anthropomorphic terminology for Fancy Matrices, is a major source of grief here.
It’s not the case I’m making, although ‘training’ is not among the worst offenders here. We train plants too.
Anything this deep into the uncanny valley can earn an impersonal pronoun, surely. I would find it distasteful to tell Codex it puts the function in the signature or it gets the hose again. I even say please.
I’m not quoting your weird rant. My reply to it would be very off topic. Stay angry, I guess. I do not share your perspective in the slightest.
But I will finish my thought. Do humans derive, by right of licit possession, the freedom to mulch text up in a big ol’ neural net?
Damn right they absolutely do, sir. That which is not forbidden is permitted, that’s called common law. I have seen no reasonable legal interpretation which would forbid it, your rant offers none, and therefore.
Fields of endeavor.
My thoughts here:
First, if you open source it and it’s useful in any way, it’s going to find its way into the training data. Even if it’s only available because you email it to someone directly, eventually someone will commit and push it somewhere that will get into the data pipeline.
If you aren’t good with that, maybe just don’t open source it.
If it’s potentially worth any money (AND you’re interested in pursuing that), don’t open source it.
1 Like
Training a neural network is a human endeavor. I don’t think this point can be successfully disputed. Them GPUs do not manufacture, purchase, installl, or run themselves.
But I’m not really paying for GPUs, am I? By using the service, I’m paying for all of the human knowledge and endeavour encoded into the weights running on those gpus. GPUs aren’t the service.
Is it insulting? Isn’t the entire point of AI that it reduces human endeavor? And otherwise why would we be investing nearly $1 trillion into these things? Why take offence at the very value proposition of AI?
I’m not quoting your weird rant.
I guess that’s a way of avoiding having to address “For-profit companies supported open source software because it commoditized their compliments, now they have figured out how to commoditize creativity and human endeavour.“
Is it really so weird to worry that AI is dismantling the economic infrastructure of innovation?
1 Like
I don’t understand. The point of all new tech is to increase human endeavor, saving money on one thing and increasing spending on another. Disruption (which can be very negative) is not the same as decreasing endeavor.
2 Likes
Good point, I think what I’m trying to say is that we’re increasing the outputs but not the outcomes, and in fact I think we’re decreasing the outcomes.
2 Likes