Ardour development and AI

Correct me if I’m wrong (I’m not a programmer at all), but I’ve read elsewhere that Linus’ “but an LLM is like a compiler” claim is complete bullshit, or at best completely misleading.

Well, Linus Torvalds is not a god or an oracle whose pronouncements are always 100% correct. He is allowed to say bullshit once in a while :slight_smile:

About AI, I don’t have much to say. It can be useful for very specific stuff when you hit a road block in your s/w developer’s work, but relying on it for an entire project from arch, design, imp, testing, etc is definitely not a great idea, even though my company’s management has recently told us that we should bring AI to the whole development cycle rather than just implementation. That’s utter BS to me and I am free to do whatever. As a side note, we are each provided with a token budget that allows any dev at work to use on Claude, Gemini, etc. For some reason, I feel a little pride that I am the only one around that has so far used absolute 0 token. I don’t even know how to access these tools. It’s completely wasted money on me … and how many hackathon projects at my work are based on AI agenting and what not. I avoid those like the plague.

1 Like

Personally I think its inaccurate - I mean we could say that a compiler is an algorithm for generating an output and we could say that an LLM is in essence a (very complex) algorithm that also generates an output (in this case code) - and we could debate the semantics of it forever, although (while I have written simple compilers myself I’m by no means an expert on the inner workings of current ones at all) you could say that it’s possible to understand deterministically the way a compiler arrives at its output, whereas we don’t know how LLMs do, other than in a broad sense.

I prefer to think of generative AI (and by extension an LLM) as a service, like employing a developer to write the thing you want, whereas a compiler is a tool which allows you to make the thing you want.

BINGO. That’s the thing one doesn’t get from using this tech. I tried this stuff years ago, thinking the same way that many do today, that it’s just an assistive technology. Now I am proud to be called a reactionary Luddite :slight_smile:

Some of my experiences:

  1. I took about a week of prompting and sorting using ChatGPT until I had a song. It looked great on paper.
    When I sang in the studio it was horrible. I felt awful and the words just didn’t go through the lips right. I started changing the lines by hand, although I left one 4 line stanza that I thought was beautiful - but there was a nagging doubt. I used a whole bunch of different search engines to try and track down if it had been lifted. About half a day of hacking and I finally uncovered where the lines came from - an obscure Swedish metal band, various bits pulled apart from different songs and resynthesized.

This has ruined the entire song for me. I feel nothing but shame, and refuse to release it.

  1. An album I released in 2022. I used AI to replace some original album art that I spent days on, with something I had spent a few hours prompting.
    In 2024 I began teaching full-time (as AI has killed my jobs as a songwriter and language translator), and the kids at the college were like: “OMG that’s AI. That is SO uncool. I hate it!”
    Again, there was no pride, just embarrassment. Still tracking down some of the singles off that album and replacing the cover art with the original. The kids are onto this caper.

  2. Ditto, with a t-shirt run I did. I spent days using AI to turn my band into little characters. Not long after, I came across a friend who was a great graphic designer and artist. She was complaining how AI had screwed her job and she was unable to feed her kid.
    Again, no pride, I just felt shame that I didn’t spend a few thousand pesos and employed my friend for a week. I keep one of those tshirts stored away, but there is zero pride. I can never wear it.

I now actively avoid LLM everything, even if I have to forgo some cool program, or spend more time on a piece of art or writing, or lose a job. I feel so much better because of it.

7 Likes

I tried chatGPT on lyrics a while ago and gave up quickly. It sounded like rubbish to me, both in English and in my (mother tongue) German.
I know someone who seems to be more successfully with lyrics on chatGPT - I think he is going back and forth using chatGPT more or less as kind of starting point. (The human brain seems to be better in improving existing things than in creating new things).

For AI on Ardour, I think I could live with AI generating stuff that is like a general library like GUI elements and the like. Things that really don’t add that much of a value to the product core functionality.

Never in life would I use an AI assistance for lyrics. Writing good lyrics (at least good to me) is a challenge I really like, the lyrics have to be musical and flow with the instrumentation of the music. I always start from some scenery in my mind that I am trying to describe. I’m the type that write lyrics after the music and always have an idea how I want to sing it beforehand. The music gives me the mental picture and the description into words that follows goes through days and days of refinement until it blends with the music. AI has absolutely no place in this process.

2 Likes

And I think this is the key point, there are aspects to each step in these processes that people enjoy, and steps they don’t. The best answer is to work with other people and collaboratively create, but whether those people are available to you or not is a different topic. LLMs are a stopgap/bandaid for that, but yes obviously human is always going to be better.

I don’t have a problem with what people are calling generative AI (I might have a slight problem with what I consider a mislabel of it of AI vs LLM) but I do definitely have a problem with “the morally bankrupt way in which it is being implemented and deployed.” definitely agree with you there. I just tend to believe we CAN do better and in a way that is not NEARLY so destructive, it is just the use of it doesn’t dictate whether that better happens or not, that is where government needs to step in sadly.

    Seablade
1 Like

I think the discussion here is interesting, but a little short sighted. We talk about one software and its possible use of AI and we do not see the tsunami that is already on the way. It will be a fundamental change just like the invention of the microprocessor was.

Try to imagine the world 5 - 15 years in the future. In that time-frame there will not be many projects or operating systems left that are untouched by AI. There probably is nowhere to run anymore if one wants to boycott AI. The change will penetrate the whole society and economics will the driving force just like it was in other technological revolutions before it. It just makes economic sense to hire one programmer and make him 10 times more productive with AI than to hire 10 programmers. It’s simply cheaper and this will drive adoption.

We do have the mouse, keyboard and graphical interfaces only because computers until now have not been able to understand natural language. When it does in many cases it does not make sense to click around a GUI anymore. You just use natural language just like you would if you were instructing a human to do a thing. You guide AI to manipulate your data using a program that has the capabilities needed and if one is missing AI will write it on the fly and make use of it. There is no more any sense to study where you need to click on the GUI and in which order to get things done. The GUI is needed mostly for visualization of the data.

I suspect about 80 % of Ardour users do not care if AI is used in development or not. They are here because they needed the tool that Ardour was able to provide. Then there is the 10% pro AI crowd and the quite vocal 10% against it.

Competition may also hurt Ardours financial model and the competition will likely use AI to accelerate development of competing products. Being left behind is also a great risk here.

I respect opposing views, please respect mine and keep the tone of discussion civilized (anticipating some flack here).

4 Likes

And, have you thought about privacy? I mean tech companies obviously wouldn’t eavesdrop on your private conversations or send / store audio clips unencrypted, would they.

If I need to tell something the kind of editing I need to do before starting mixing guitars, bass and drums just to have everything leveled, try and fix the occasional mistake I didn’t notice about (in my case, guitars (usually quad-tracked) and bass need to be a one take) and make sense of it all before adding the first plugin I swear that before that I will sell all my instruments and buy me a theremin to play alone at home.

That already existed. Is a mainframe terminal with a microphone. Even if it is local is a privacy nightmare.

The constant message that in the future with a couple of developers all will be good assumes that the price of AI conssumption is going to be as of today or less which is plainly unsustainable and not realistic.

I personally know a CDO that took that route a year ago promising costs cuts in the near future and now he has fallen back to… Hiring in India.

Ah! The wonders of the corporate world…

Set your sights a little further. In the 1 - 3 year time-frame prices will go up, but progress will not stand still. Take a look what the world looks like in 20 years. Competition and technological progress will drive prices to the ground and AI will be integrated in your devices and maybe some toys. No need to use those data center models anymore. Take a look 50 years in the future or 100. What will the society look like when collected human knowledge / wisdom in the form of AI can be made to tackle hard problems, those that are currently out of reach.

I thought I’d drop this very recent video by Red means Recording on the subject.
Jeremy has used LLMs to generate code for audio projects in the past, but he has decided to move on from that and he explains why.
He’s a musician, a prominent member of the synth hardware scene, in direct contact with most hardware and software manufacturers in the pro audio world. I think his opinion is worth considering:

This kind of bad behaviour of tech companies is more a result of closed source and proprietary licensing than A.I. If tech companies want to ride rough shod over licensing, then they certainly shouldn’t be able to benefit from them on their own stuff.

That is extremely debatable.

AI data centers are driving memory prices up and up, to the point where people begin to tinker with (still relatively) cheap ESP32-based boards to run pretrained and/or simplified models. Which, let me tell you, are very imprecise and painfully slow for obvious reasons.

Well, original LLMs will become more heavyweight, and MCUs cannot rival servers, desktops, or SBCs. The ongoing trend won’t be linear, but we don’t really know where that curve will go.

If you are sure you know how this will play out in 5 years time, let alone 20, I have no idea what you are doing here. There is a very short supply of genius investors who can time tech markets, are billionaires in disguise, and hang out on FOSS forums at the edge of interwebz. I don’t recognize you as one of them :smiley:

I do agree with you that LLMs (or other forms of AI tech) will be our future. Sometimes a welcome one, sometimes shoved down our throats.

1 Like

I wish I was a millionaire in disguise, but I’m not :grinning: Those investing in AI companies now may either win big or lose their money. I think those companies will be very profitable for some years but soon the market will be flooded with companies selling models and the competition is going to be fierce.

In the 70’s we had monstrous mainframe machines that were so expensive they could only be used as timeshare service. Fast forward 50 years and we have the computing power of tens of thousands of those machines in our pocket as a phone or tablet. And a smartphone is 10 000 times cheaper than what a mainframe used to cost. Now many people use AI the same way mainframes were used back then.

Technology will be optimized for cost and performance as far as the physical limits allow. Market competition takes care of that. That is why I think we may see history repeating itself here. But only time will tell.

1 Like

Now, here we go, all Technology ever Invented made us Weaker. AI aims to make us stupid, (Who in this World might have advantage from that)?
speaking for my self: I am Lazy, weak and stupid enougth, I guess. :slight_smile: Btw.: Who killed John Henry? :smiley:

Who would have thought that with all the world’s knowledge at our fingertips, we’d be more stupid than ever… :slight_smile:

4 Likes

Knowledge without will and wisdom is indeed quite vacant…

1 Like

It seems they sensed it early …