Auto complete, as far as I understand, is not based in machine learning. The Linux devs knowingly allowing such things into the kernel does not validate its use, as all the inherent problems with “gen” “AI” extend to LLMs as a whole. If anything, its troubling to see the open source OS endorse deeply harmful practices bankrolled by the same big tech companies many people are trying to escape from when running Linux.
1 Like
“Auto complete on steroids” would be e.g. I write void add_new_midi_track ( and the LLM fills in the rest, incl arguments and implementation 
1 Like
Sorry, what’s the confusion here? If I “generate” an image, even if I add something of my own, that does not make it my own in any way. Whatever I added is now tainted by the “generated” the image from which I was working. It can’t be my own creation no matter how much I change it.
Fair, and like, I understand how something like that could be useful. But is the tiny bit of convenience really worth all of the baggage that comes with LLMs? I’m trying not to restate all the details in my replies, as that’s annoying and stressful for me. I’m troubled that the usage of something like this is even on the table for this project.
It currently isn’t for Ardour, but the Linux kernel does set a precedent to keep an eye on.
1 Like
That isn’t a description of music, or writing, or visual art.
Why should it be a description of source code?
A derivative work is still your own work, it’s just built on a foundation rather than starting from scratch.
This is very common in classical music where you take a theme composed by someone else and build on it, or re-orchestrating an existing piece of music, giving it a whole new character.
Not to mention pretty much every jazz “standard”, which began life as a show tune or pop tune 1-3 decades earlier and were transformed by the players, sometimes unrecognizably, but often in ways that left the skeleton of the song intact, but nevertheless something very individual to them.
2 Likes
Funny your nickname relates to the link …
Astonishing enough, one of the most influental skateboarders, Rodney Mullen, kinda covered the topic , minus llm of course, but “what is your work” or “what is creativity”.
Not trying to make a point, agree or disagree, i just like the speech for several reasons, and perhaps others like it too. I hope so
“Pop an ollie and innovate”, Rodney Mullen:
I misspoke in my last reply. I shouldn’t have said it doesn’t make it my own “in any way”, as if the “generated” portion absorbs my own contributions into itself or something. There is a kernel (ha) of truth to the idea, however, as derivative works can only be built off the efforts of other people. While LLMs are trained on human creations and can only output what they’ve been fed, it will always result in a garbled mismatch of potentially hundreds of works, most of which are stolen, none of which can be attributed to. To use an LLM is to erase the history of everyone’s work that is being stolen from, reducing most every attribution to “the human race, sometime, somewhere, who knows”.
1 Like
How do you feel about (historical) Stack Exchange ?
Human beings responding to queries by other human beings? Engaging with each other instead of wasting copious amounts of resources and energy interacting with a theft machine that will never provide reliable information? Knowing that if someone consistently gives horrible answers, there’s at least a username to hold accountable and take action against?
Accountability? With Stack Exchange? You are kidding, right?
Sure you could down vote, but that’s about it. And much of the information on it aged as well as uncovered milk.
Cheers,
Keith
OK, so there are (at least) 3 interwoven, but distinct issues here.
- Profligate use of various natural resources, including energy
- Wanton disregard for copyright laws
- Lack of attribution for previous work that plays a role in new work
They are not the same, and I’m not particularly concerned with #3 because my view of human culture tends to emphasize “stealing” (as in “great artists steal”) and downplays the idea that human beings own culture even when they create it.
#1 is critically important, and #2 is also concerning. I do not believe, however, that by providing an MCP interface to Ardour, that we are necessarily engaging with #1 (because you could use a local instance). I also think that the intellectual and practical questions surrounding #2 are complex, and not sensibly reduced to “they read everything and now they spit it back out”.
3 Likes
Well, I pushed the new code to PR #986. Take a look.
Full disclaimer: I supervised the whole process, but all of the C++ code was written by Claude; I wrote the Lua code myself last year.
If it’s rejected because it’s AI-generated, no hard feelings. I fully expect that, based on the stated policy. I’m just concerned the free software community, in general, is rejected a breakthrough technology.
Plus, I’d like to feel that I’ve contributed something useful to Ardour, however small it may be.
I never even heard of Stack Exchange before today, so apologies for not having a history with it and having to go off of the Wikipedia page. Regardless, I was obviously referring to the fact that moderation could be applied to actual users instead of some bot.
I also do not think people own the culture they create. At least, not exclusively. My previous comment was less about the MCP integration and more on the idea that something LLM “generated” could be used as a starting point without resulting in a work that is tainted. It’s true that people take subconscious inspiration all the time, but that is still rooted in the individual, their experiences, tastes, and whatnot.
Personally, if my work influenced someone else, I would be proud of that. If my work was instead thrown into a computer then regurgitated at random into somebody else’s project, I honestly don’t know what I’d feel. It wouldn’t be great.
As @plane pointed out, LLMs are pretty intertwined with fascism. I’ve tried to stay away from the term because of how loaded it is, but there’s no getting around it. They said it better then I can, honestly, so I think I’ll just leave it here.
Bots are users too. They have accounts on something like Stack Exchange, and are equally subject to down voting or moderation. The are indistinguishable from human users, in that respect.
Arguably, just like this community: I could be an LLM bot (or a human driving an AI for responses) and you would probably not know. For the record, I am neither of those things (but that’s what a bot would say…).
If you are comparing Stack Exchange with, say, a straight AI chat (e.g. ChatGPT, etc.) then you may have some point. But the difference there is in the interface/environment and the lack of pushback from other users.
But, as an intellectual exercise, if you took a Stack Exchange like interface which was populated with multiple, competing, LLM model bot “users” with different briefs and emulating human dynamics, then I suspect it would appear very similar to historical Stack exchange and would offer much of the same value to human readers/users.
I’m not defending LLMs here. Nor am I attacking them. There are very real concerns, which Paul and others (including yourself) have highlighted.
Adding to the issues others have raised, I genuinely believe that for certain tasks (like coding) the reliance on scraping human data is going to be the downfall of LLMs in the long term: if humans stop producing content, then LLMs will eat themselves, and that has been proven to result in even worse slop than we get today.
I also, honestly, think that LLMs will spell the end of copyright as we know it. It’s an artificial thing anyway. The problem is, industries and careers have been built on it, and the disintegration of that is going to be messy. I fully support the right for artists and performers to be paid, but I think that the status quo is broken and LLMs are demonstrating that.
For decades now, humans have exploited talented (and no-so-talented) humans for profit. What is happening now is both an extreme version of that - talent-less billionaires trying to score their next billion by screwing both artists and users (Suno, etc.) and an erosion of the classical artist development/distribution system.
That system has IMO largely been broken for decades, with labels pushing formulaic, cheap, marketable crap and “image” over real artistry. the only real difference is that they can do it with AI rather than humans. The downside of that is that is that removes career opportunities for artists. The upside is they will no abuse and bankrupt gullible young artists (which is the basic MO of the record industry). IMO almost all of the really talented, interesting music in since around 2000 has come from independent artists marketing themselves.
The elephant in the room is the massive streaming sites, especially Spotify, which is still exploiting old-school consumption habits whilst surreptitiously replacing humans with AI.
I suspect (and hope) there will be increasing recognition of this and pushback/abandonment of these platforms. After all, people follow artists because they like to have a connection to them. If they start to realise that there’s no actual artist behind the curtain, and no actual art in the material, then that relationship collapses.
I could easily eventually see Spotify being replaced by a “replication” service (I’m thinking Star Trek replicators here) where you can create AI generated “musak” to your hearts content from a prompt for a very low cost per month, rather than paying Spotify (why get Spotify to do it and sell you it when you can do it yourself for far less).
People will, IMO still make music and share it, whether on communities like this, or venues. You can’t replace live music with AI (at least, not very well).
And yes, that’s going to affect people’s livelihoods in the way we currently run societies, but the way we currently run societies is also an artificial construct, largely invented/shaped in the industrial revolution by those with wealth and power. It can change (think Star Trek again…).
All of this is to say, whilst AI and LLMs specifically can be considered bad in many respects, they are not going anywhere any time soon. Railing against them wholesale (including anything which isn’t LLM but contains potentially LLM enabling technology, like Ardour) is tilting against windmills. It’s not going to go away. That doesn’t mean we should be dropping our trousers and bending over the barrel for the tech-bros who are pushing them either!
Cheers,
Keith
Why does everybody repeat this all the time? It seems to me that the chatbot craze will go away or at least decrease pretty soon (starting end of this year?), now that the investors are beginning to ask for profits. This technology is not sustainable. The tech-bros know that, and now want to go to Wall Street, hoping to distribute the losses to small investors, pension funds and so on.
The adopt-or-be-left-back narrative is fabricated by the same corporations trying to sell us those chatbots.
That’s not what I was saying. I was saying that LLM technology is not going to go away, just because some of the more enlightened amongst us rail against it. Just as proprietary software didn’t go away because of open-source.
Yes, it’s a bubble, and much of that bubble will pop. not just i financial terms, but in technology sustainability terms (as I understand it, the LLMs have already pretty much reached the limit of scraping all available content on the Internet and are running out of material - I mentioned them eating themselves…).
The other constraints are available power and water. In the UK, for example, the requested power for new AI data centres is more than the current power consumption of the whole UK. That’s simply not going to happen in the next 10 years.
But the technology will survive and persist, just not in the form the tech-bros are currently pushing. And it will have significant impacts on society.
Putting your head in the sand and expecting it to go away, just isn’t an sensible option!
Cheers,
Keith
1 Like