QUick immediate response: there is absolutely no way that we’re ever going to manually review the 16K LOC patch that I mentioned. We don’t even have the expertise to do it. However, since it is GUI toolkit technology, we assume that we’ll get bug reports if it is broken, and we’ll fix them (likely by hand). This particular patch is totally isolated in its own library, and Ardour could be built without it.
We would never take that approach to “actual” Ardour functionality (GUI, DSP, whatever).
As a mantainer of a relatively popular open source project myself (nowhere near as popular as Ardour, of course 
), I think your analysis was very reasonable, and did resonate with me for the very same reasons, especially considering I’m pretty anti-AI myself at the core (I very much despise its use in arts, for sure).
In my project’s case, I decided to be somewhat tolerant of AI contributions (as more and more did start to arrive, and many of them were indeed unmistakable fixes to security vulnerabilities), but I started asking people to disclose it as part of a new pull request template, and to describe (in their own words, rather than just the wall of text AI produces for patches) what the change is for, in order to make sure they understand the code and implications. I personally don’t use AI tools to produce code myself, if not for very limited cases (e.g., web workers stuff for a demo, recently, as it’s an area I know very little about): I prefer to only occasionally poke them as a sounding board, when I’m hitting a wall and may benefit from figuring out if I’m, let’s say, on the right direction, or what I may be doing wrong.
All this to say that I personally feel you should do whatever you feel makes it easy to manage the project, in a way of course that doesn’t go against your principles. Not everyone understands what it actually means mantaining such a big project, and very antagonistic as I am to AI in general, I think it would be myopic of those who don’t know to impose their own principles on you. Reading there may be people who may decide to stop sponsoring the project just because of the fact you may use some AI to help you work makes me sad, as we should help mantainers keep things, well, mantainable, not put even more burdens and crosses on them.
So, as far as I’m concerned, I’ll be ok with whatever road you choose to take!
Well, as @rasmus said, points 3, 4, 5, 6, 7 and 8 of the proposal are not being addressed, and 1 and 2 only marginally (i.e. there is no clear legal status, especially in the EU). Thus my opinion is the project should remain LLM-free.
I don’t know how the ‘structure’ of the financial contributors, but given that one can get Ardour from a Linux distribution for free, I would argue that many (most?) are contributing to the project, and not to be able to download a precompiled binary for a mere 1 USD. Just guessing, based on the impressions I get from the forum.
@GMaq @josander @LinuxGuitarist @GhostsonAcid @lminiero: are you really putting on the same level saving time for tedious tasks and the destruction of the only environment where humans can live? Also the rise to power of nazi-tech-bros?
Also, just because the chatbots are being constantly forced upon us, it does not mean they are “inevitable”. In fact, it means the opposite: nobody forced us to use a washing machine.
@paul and the other developers: I am not pretending to decide how or how much you have to work. I am writing this because you asked for it. There you go 
Thanks for this opportunity.
are you really putting on the same level saving time for tedious tasks and the destruction of the only environment where humans can live?
Every technology has environmental impacts. Hell, you’re posting on the Internet, which is one of the most energy-hungry technologies in human history. Environmental impacts by themselves are not a valid reason to reject a technology (though they are a valid concern).
Also the rise to power of nazi-tech-bros?
Such a ridiculous, politicized, and inaccurate assertion that I am not even going to justify it with a rebuttal.
Also, just because the chatbots are being constantly forced upon us, it does not mean they are “inevitable”. In fact, it means the opposite: nobody forced us to use a washing machine.
If you prefer to hand-wash your clothes or not use AI, you are free to do so. Just know that not everyone shares your view, and many (most?) will choose the more convenient option.
Let’s compromise.
How about: a 50% fast team? 
I absolutely agree. But look ahead further: a local, at-home, Ardour-trained, AI assistant (i.e. super autocorrect) could potentially remove a significant chunk of the ethical problems. … Plus, doesn’t Paul live in the barren desert? ~He could buy a solar panel or two to power it. 
Suno and anything that replaces the actual art is the absolute worst. But as you pointed out, @paul and @x42 are treating this with care, -and explicitly understanding things like: “Ardour is a tool, not a final product”…
There is no way in hell that Ardour would ever stop being maintained because Claude Code was down, haha. Again, it’s not like Paul’s out here with an ‘I 
AI’ hat on, ready to become the next DHH.
Mixing is art. There is zero doubt about that in my mind. “Editing”, “Mixing”, etc… –these are vital to the shaping of the art, as vital as the performances captured. So in my opinion, their lack of enthusiasm to include any AI-assisted “mixing” features in Ardour, etc., is totally reasonable and appreciated. -Leave ALL that to the SLOW HUMANS.
Well, I suppose I had that more front-facing code in mind, yes.
Again, one day soon they might be able to gain assistance from a 100% local, FOSS-based LLM setup. -Totally possible in the foreseeable future. Data centers are indeed a cancer upon the Earth. I hope they can be made obsolete sooner than later with lighter, local tech, on par with what’s happened for computer processing over its history.
Mixing is art. There is zero doubt about that in my mind. “Editing”, “Mixing”, “Mastering”, etc… –these are all vital to the shaping of the art, as vital as the performances captured. So in my opinion, their lack of enthusiasm to include any AI-assisted “mixing” features in Ardour, etc., is totally reasonable and appreciated. -Leave ALL that to the SLOW HUMANS.
Mixing can be an art, but for some it’s just a chore. I don’t particularly enjoy mixing, and I can’t afford to hire someone to do it for me. I believe that it is totally reasonable to want to automate the mixing process if I don’t enjoy it. If you enjoy it, you are free to do it the old-fashioned way. A mixing assistant in Ardour wouldn’t stop you from doing that.
Hi,
I use ardour for more than a decade by now, and I really appreciate the way it has evolved since then. I think this DAW is a masterpiece in FOSS world, and for that I thank you so much. It had assist me in so many and various projects.
When I was a teenager I switched to linux to escape from the capitalist hell of proprietary softwares. In the same move I discovered Ardour, which offered me the possibility to stay in that way, and gave me a wonderfull tool to continue.
Nowadays, the capitalists I fled in my youngness prove to turn into fascists pricks, and are trying to force everyone to come back into there dominion through IA.
I also know the work to maintain Ardour is huge and endless. But to me, the sense of using it isn’t a new tool or feature, but in the possibility to continue using Ardour, the farest from IA hell. Especially when any moron can ask suno or I-don’t-know-what-IA-crap-tool to generate in a few secondes an amazing tune that would take hours of work to any human, I trust in the significance to ignore all this shit and just do what we love to do, at a human rhythm, spending to much time for this, but feeling free.
I remember so many talks and situation where people, friends, colleagues didn’t understood my obstination using ardour or other Foss, instead of good all cracked logic, pro tools or ableton. But in the end it as always been about one and only thing : freedom.
So, yes please, keep us the farest possible from IA features, and all this crap.
About the development of ardour, I won’t give you any lesson as I don’t understand such wides parts of it, but I still hope you can continue without IA.
Thanks
It can be tedious, but you can do it. You can also find a local artist who will do it for a small fee or even for free, just to grow their portfolio. There is already so much unemployment in the arts, let’s not deprecate artist even more, please. For you it’s a nice tool, but thousands of other people will use it to avoid paying humans. And without humans, all mixes will be bland and boring, because machines can’t innovate (and even if they could, it would be meaningless without the human element).
Part of the dilemma here is that there are a number of coding tasks that can be tedious too. Why would we treat them differently than, say, mixing?
One answer might be: find me the human (anywhere!) who will write the 16K LOC required to implement <left unstated for now>, work that will bring precisely zero new features to our users (or possibly, arguably, just one). If such a person cannot be found, is it “just life” that said work will remain undone? I’m not saying that the answer ought to be “yes” or “no”, but I am pointing out that I think “it’s complicated”.
There is already so much unemployment in the arts, let’s not deprecate artist even more, please. For you it’s a nice tool, but thousands of other people will use it to avoid paying humans.
Nobody is entitled to a job, especially in the arts. And we should not retard technological progress simply because it puts people out of work. I have no issue with technology taking the place of humans if it makes it easier or cheaper for myself or others to get things done.
And without humans, all mixes will be bland and boring, because machines can’t innovate (and even if they could, it would be meaningless without the human element).
Debatable. A large chunk of mixing is just getting the balance right and dealing with frequency clashes. Those are the parts I find boring, and wish to automate. I probably wouldn’t let an AI completely mix my track, but I would definitely let it do the grunt work.
And of course, I would still be writing and performing the song itself. So there’s a human element in that.
My concern is - why use it if the code cannot be GPL licensed? Let’s say you stumble upon a great idea for a new feature, vibe-code it, and it turns out bugles and production ready. Wouldn’t it be dissapointing to not be able to GPL license it?
Well, I suppose it’s a mixed bag, as always.
I mean, even a compressor is an automator of manual fader riding.
-Tedious!
Everything’s ultimately a balance, as cliché as that reads… 
Some things you should lean into the automation.
Some things you shouldn’t.
I suppose that’s up to each of us to decide, hopefully as ethically as we can.
The current legal situation in (at least) the EU and the US is that if a human “plays a sufficient role” in the generation of the code, then they may copyright it (and if they copright it, they may obviously choose whatever license they wish).
The dilemma is that what “sufficient” means has not (yet) been decided by the courts, anywhere.
…woudn’t it mean that you’ll have to maintain separate versions of future Ardour releases? The standard one plus the one with AI generated code additions? I mean, imagine you come far with AI generated code, and then after all of it, the courts rule out the possibility of proper licensing of newer Ardour versions?
I think AI would be a very good way to add more resources to Ardour development. So go for it 
Also please follow what Linus Torvalds does with the Linux kernel. He has all the same dilemmas as you (licensing, contributions, etc ). He will probably be the first to come up with ways to deal with these so that you don’t have to invent the wheel all by yourselves. Also if there is any legal trouble the kernel will be the first target and smaller projects will go under the radar.
I think you could filter out sloppy contributions also by AI review that makes sure the code follows your standards and coding style. You could create an AI “skill” that defines all this and users could run their contributions against that before making a pull request. There is no need to review all code manually, let AI do most of the work and review the code only after it is up to the standard.
In my opinion AI will be a great open source accelerator. Most projects struggle with finding resources and now there is a way to overcome that.
This is why we will not accept LLM-generated code in cases where the person who was involved in the generation will not or cannot claim copyright.
Well, your idea is great, but, at the same time, doesn’t the court have the final say here?
Let’s say you yourself generate a new feature with assistance of Claude. You try to GPL license it, and the court says no - they say it’s vibe-coded and similar feature alraedy exist in xyz proprietary DAW, even though you’ve spend a great deal of time&tokens developing it. Then what?
Even worse…imagine that happens 3-4 major versions down the development timeline?
Then we’d have to remove it. But I think that scenario is less likely than a court saying “you didn’t do enough, that was machine-generated and it is not copyrightable”
So, this should be a fun one to jump in. Basic background on my usage currently:
I have started really experimenting with LLMs about 6 months ago, maybe less. I started looking at code generation in LLMs about 3 months ago. This came about when I realized, I had multiple ideas that would help me teach better, or rather help my students be able to explore and learn better, but I had no time to be able to implement it with the limited coding knowledge and experience I have (See above list of what I do).
What came of this, was generating an ear training piece of software, that I could tune to meet my specific needs, including integration (To a limited extent) with the LMS used by the university to sync grades and students etc. and assign tasks to complete. This came after trying a couple of different paths to accomplish this. I had a budget I was willing to spend on this, but it wasn’t enough to hire a developer to assist me with the programming, and explaining some of my DSP requirements doesn’t mean that a random developer would understand it well, and even the initial quotes I got for completing this task would likely be far lower than the actual cost.
Instead I started vibe coding the tool and within a month had a workable version of it I deployed to my students. I have remained under budget, but not by much, of what I had to complete this tool, which my budget was in the 4 digit range, to complete a v1 (Technically I am working on v0.8 right now) with plans already developing for a v2 of the tool. Vibe Coding allowed me to create the tool for the betterment of my students and for my own use of time in the future, that would not otherwise exist.
That being said, I HAVE had to closely monitor it. It is not that this is a major time saver entirely, and while it did cost less money than hiring the developer, so technically saved money, my time is obviously worth something as well, and I may or may not have had to spend as much time overseeing it as I would a developer, so debatable how much savings there is overall.
I do think there is value here. I think the points about the environmental impact are 100% accurate and need to be addressed, but I also do believe that the environmental impact, while directly attributable to the growth of data centers especially, is a larger issue than just LLMs, it just happens they are the driving force currently, but I also suspect there will be another force to drive it if LLMs didn’t exist. Remember, data centers have existed for decades now, and have been growing in that entire amount of time, not at the speed lately, and the stupidity I have seen with some projects, but they definitely did. So I agree with a previous poster that the issue here is more than just ‘LLM’ and a societal and economic situation that LLMs happens to be the foremost driver of currently.
It is also worth noting that when people talk about the environmental impact, they are talking in particular about the development of new models, more so than the use of current models, which have vastly different power requirements. There is also a lot of misinformation about data centers and environmental impacts, which there is a VERY real environmental impact happening, and an economic impact in my area, but some of the reason I see people throwing out aren’t actually accurate or are very specific edge cases. All of these issues NEED to be addressed, but that addressing needs to happen at a governmental level as well.
I am in favor of finding a way to incorporate this technology in a reasonable way. I think this discussion is the beginning of it and overall I trust Paul and Robin both to weigh things out carefully before moving forward. I am not sure about the legal ramifications yet, and doing this in a way that keeps the code able to be licensed properly as open source is important. In fact I think this is even more examples of why open source development can be so important, and is bringing the strength of this to a larger and more direct impact for people. I do believe LLMs are going to exist for some time, I also do believe there is a huge bubble around that industry that will pop at some point, but I do believe LLMs are changing the way we work in similar ways to what computers did back in the 80s and 90s and we need to be able to adapt.
I am sure there is a huge amount of thoughts on this topic that i haven’t spoken to yet. But short version I am in favor of the approach Paul and Robin seem to be taking.
@Ljuba Features can not be copyrighted. Only the exact code can be. You can implement the same functionality that exists in proprietary software if you write the code differently. This has been done for example in the case of creating a copy of the first IBM bios for the clone-pcs to use.
Paul means that if AI code can’t be copyrighted then it may not be compatible with the GPL and can not be put under it. Linus has this same problem (times 1000) and he seems to somewhat embrace AI.
Another thing is will anyone ever complain if AI code is put under the GPL license. There is not much to gain bringing that to a court, is there ?