In the great web tradition of harvesting the vast body of other people's work in the large[1] and shoving it through huge amounts of computation to wring out a nickel's worth of value that will eventually manifest in some good-paying SWE jobs, a rich executive class, and a whole lot of shareholder value and inevitably mutate in another goddamn ad-serving platform.
[1] Ha, the poor millions of dumb minions who put their work on the web thinking it might be fun for others or garner themselves a small following, they didn't check the terms of the EULA!
These kinds of discussions always leaves me wondering if people consider how actual humans learn their craft, constantly studying and mimicking others. Inspiration is to use existing experiences however mixed together, while originality comes from an input or an experience that others have yet to use.
"Write a sad song about the MIT license" is certainly such new input, and if I was commissioned to write the song it would be based on inspiration (i.e., "use training on") music I have heard or studied. And yes, none of the musicians I have listened to or have studied will benefit from the endless money fountain I'd acquire from composing such song.
> In the case of a human studying, a person puts in effort and gets rewarded for their efforts.
When someone needs something composed, they don't learn how to write music. They pay someone else the bare minimum, e.g. a few bucks on fiverr. The person will spend the least possible amount of effort to try to make their life go around with the little money they got.
When you then use an AI model, the work done for those five bucks is replaced by work done for almost free.
Neither the person you would hire or the AI credited those who created the material they trained on.
> When you then use an AI model, the work done for those five bucks is replaced by work done for almost free.
In other words, you pay a few cents to big tech for a generator that only exists thanks to work real composers, singers etc who now get the grand total of 0$.
As opposed to paying $5 to a singular composer that only exists because of other real composers, singers, etc., as they studied the craft. On the other hand, the easy access may cause new types of artists to appear by lowering the bar of entry, or just make custom music more generally available and more widely used as the price makes it a commodity.
We also stopped hiring computers (the occupation) and instead pay big tech companies which made computers (the device) available to everyone. And we stopped hiring people to do dangerous manual labor as companies started selling machinery to automate it. Markets change.
> As opposed to paying $5 to a singular composer that only exists because of other real composers, singers, etc., as they studied the craft
Yes.
> computers
> dangerous manual labor
If people are not in danger and don't have to do mechanical work, it's one thing. If composers stop composing original music that was used for training current AI because they don't get paid anymore then the field stagnates. Same for writing and everything else
People doing mechanical work made art through physical objects. Woodwork, pottery, glassware, you name it.
There are now far more options, both on the high and low-end, with the whole area being more affordable. The quality of most products also arguably went up, as factories beat handmade goods. And yet, if you want custom artisan goods, you can still pay a woodworker for it at more or less the same cost as you would have otherwise, as their labor costs are a function of time required and local living conditions.
In some cases, those workers were the ones to automate, benefiting from the assistance - woodworkers using CNC mills and laser cutters even for handmade goods, or composers themselves can use the AI - to speed up their otherwise fully manual work. It benefits the majority creating the demand, and tends to improve the craft overall.
> ... then the field stagnates.
A market that is not changing has already stagnated.
Dangerous manual labour like mining or tunnel building. In woodworking the art is art but if you use a machine to copy my original design then it's the same old theft again. :)
> A market that is not changing has already stagnated.
Yep. The way art is changing is thanks to original work and no one will be making it since anything you make gets stolen for free
> When someone needs something composed, they don’t learn to write music…
Speak for yourself! There is only one thing that scares me more than composing music, and that’s paying somebody a few bucks in fiverr to do it for me.
Despite your personal fears I believe I spoke for the vast majority of cases rather than just for myself.
Although I suppose royalty-free stock music is the norm nowadays for most commercial uses, which takes it a step further, anonymizing the composer entirely...
> Although I suppose royalty-free stock music is the norm nowadays for most commercial uses, which takes it a step further, anonymizing the composer entirely...
By the composer, yes, but the composer here is the AI. In neither case did the musicians that the composer studied/trained on get asked.
And that's the point: The difference is the replacement of 1 flesh-and-blood composer with 1 virtual composer, with the consequence being the lost business of the former. The artists studied were never part of the transaction in either case.
Now, the long-term consequences for artists - e.g., reduced supply on the low end as they're out-competed - is harder to guess, but that's just market dynamics. It may very well increase supply as composition becomes more available, diversifying by allowing people with other skills or creative treats to create music that previously could not - even if the musical part is done by AI.
Not AI, people who trained AI and who use it for profit
You can learn yourself but if you use an automatic tool to bypass and automatically make similar works and compete with original authors then you're IP thief
> In the case of AI, a person puts in minimal effort to generate something that devalues the work of all the people who did put in effort.
Worded differently: people who couldn't otherwise produce skill-based works of value have had the barrier of entry lowered for that specific medium of expression, allowing for more works across a wider spectrum of skill.
It’s so bizarre when people say stuff like this. There is absolutely nothing preventing the unpracticed or untalented people from any form of creative expression. What instead people who use AI seem to want is for unpracticed or untalented people to perform at the level of the practiced and talented, but this is no net gain to anyone. Why? Because only a rare subset of people who ARE practiced and talented create anything of interest or value in the first place. What this tells you is that skill or level of performance is not the barrier, but a means through which great things CAN be achieved (i.e. necessary, but not sufficient)
Flooding the world with unpolished, unpracticed works, AI-tuned to the level of being mediocre, is a creative and intellectual dead end.
> for unpracticed or untalented people to perform at the level of the practiced and talented
This is what tools are.
Cheap digital tablets have done away with the need for expensive consumables. You can just download a different brush style instead of learning a physical technique. No waiting for paint to dry or smudged pencils. The barrier to entry for painting has dropped to a one time investment of like a hundred bucks. Almost nobody mixes their own paint, nor stretches their own canvas. Those skills aren't needed anymore.
It's possible to build very precise machine parts by hand. It's very difficult and requires great skill, so nobody does that. Some do and are admired for it, but everybody else uses precise machines to make precise parts with nearly no effort.
It's just a tool. Only difference is that we had assumed art would never be automatable.
Objectively, I don't think this is a bad thing. It doesn't change the subjective value of art any more than the average cartoonist devalues the Mona Lisa. It's just a new form of art, there will always be people mixing their own paints and stretching their own canvas, just as there always has been.
It's only a problem because in our society you either have a job or you starve. No one can afford to be an artist. Those that do tend to grind out as many pieces as fast as they can so they can pay the goddamn rent. If not for that, these AI tools would be pretty cool.
I think the bizarrity arises from the following differences in beliefs:
* That "_any_ form of creative expression" is a viable creative substitute for people wanting to create in a _specific_ medium of creative expression -- especially those that had a high barrier of technical skills required to be seen as "good enough" to share.
* That a person who has an idea for art will put in the necessary time to become proficient enough to create that "good enough" art through traditional means (IMO demonstrably incorrect), and that is preferred over that person just not expressing a lower-quality version of that idea at all.
* That those who use AI primarily want or expect to "perform at the level of the practiced and talented" (i.e. top-tier art) rather than using it to produce art they otherwise couldn't have, even at low- and mid-level qualities.
* That there is no skill or talent in using AI tools to produce art (or that the skill or talent using AI tools is meant to be a full replacement for traditional artistic skills or talents).
FWIW, I'm a long-time sketch artist and acrylics painter (~20 years). There are many mediums, subjects, and styles that I'm not good at -- and I enjoy using AI to express myself in those areas (and have also liked using AI to create songs to show to my more musicially-adept wife...). But even in my own wheelhouse (landscapes and still life), I also often use AI to brainstorm composition, perspective, colors, textures, lighting, etc. It's a great tool for experts to lean on, but an even better tool for non-artists who couldn't or wouldn't otherwise share their art.
Indeed. As an amateur guitarist, but a professional virtual machinist, I have a ton of respect for people who have dedicated their whole lives to mastery in any one particular area. To have a machine gulp down untold eons of human exertion and then barf out soulless mimicry, no matter how jaw-dropping of a feat of engineering behind it, and then mint no-talent ass clowns by the million because viral videos make an awesome advertising platform--it's just some kind of dystopian peak tech, except the dystopia is mildly amusing rather than a disappointing and jarring marginalization, flippant dismissal of all of us.
Why is this the line? Where are the complaints about people using pianos to achieve rather precise notes instead of using their own voices? They are just untalented at singing and their use of any tool to create sound is of no net gain to anyone.
This person: https://www.youtube.com/watch?v=IbUE-LxhUR8 ? They're recording and playing back on a loop! They should record full repeated playings, any use of the recording is of no net gain because it could be achieved otherwise.
Songwriters? If they write lyrics and someone else sings them the result should be cast into the sea - it's of no net gain to anyone because they did not create the sounds themselves.
Composers? Frankly pointless.
> Flooding the world with unpolished, unpracticed works
I hate to break it to you but there are a vast number of terrible works of art out there already.
> What this tells you is that skill or level of performance is not the barrier, but a means through which great things CAN be achieved (i.e. necessary, but not sufficient)
If it's a necessary thing, of course it's a barrier. That there are two barriers doesn't change that.
Even the most derivative of singer songwriters tend to use their own voices rather than a weighted average of the voices of other singers in their genre...
Using the skills they presumably developed listening to and copying other singers and studying music, with an instrument built from roughly the same instructions as everyone else.
That a person can't sound like the weighted average is human limitation (although with modern pop people do get quite close!), not because new singers aren't trying to. That of course adds variation that we appreciate, but doesn't change the underlying similarity in how acquired skill is mimicry of those who acquired it before us - with very rare exceptions.
No, sounding like the genre-weighted average of Spotify simply isn't what singers try to do. They haven't listened to that much music, they have actual preferences, they have natural qualities to their voice which they're complimented on or asked to mask, and they're trying to hit notes based on their aural perception of harmony and related theoretical principles not based on the waveforms of other songs involving singer songwriters. The fact that they literally couldn't do what NNs do even if they wanted to also seems quite relevant to the fact that they don't do what NNs do.
What next, are we going to argue that what programmers creating new programs are really trying to do is generate a prompt-weighted average of the bytecode of every program they've ever downloaded, and all that business analysis and functional spec and use of high level programming languages and expressed preferences for coding standards is irrelevant?
That's the physical limitations I referred to, which isn't something humans tend to be happy about but can sometimes end up being a differentiating benefit.
> What next, are we going to argue that what programmers creating new programs are really trying to do is generate a prompt-weighted average of the bytecode of every program they've ever downloaded
That's a horrible strawman. Do you as programmer often read and write bytecode directly?
I'm beginning to assume you're an LLM, because I'm not convinced a human would honestly try to argue that their emotional reaction to their favourite songs is basically equivalent to flipping the values of some bits to ensure that they generate music more similarly to them.
> That's a horrible strawman. Do you as programmer often read and write bytecode directly?
As an improvising guitarist (even a very mediocre one) my creative process is even further removed from an LLM parsing and generating sound files directly....
I suspect the issue here is just the assumption that LLMs are "just flipping some bits", while simultaneously putting humanity on some unreachable pedestal.
We are all nothing but a horde of molecular machines. Your "you" is just individual neurons reacting to input in accordance to their current chemical state and active connections. All your experiences, unique personality treats, and creativity you add to the process is solely the result of the current state of your network of neurons in response to a particular input.
But while an LLM is trained once and then has its state fixed in place regardless of input, we "train" continously, and while an LLM might have experience of an inhuman corpus for a certain subject, we have many "irrelevant" experiences to mix things up.
Your "prompt" is also messy, including the current sound of your own heartbeat, the remaining taste in your mouth from your last meal, the feeling of a breeze through your hair as it tickles your neck, while the LLM has just one, maybe two half-assed sentences. This mix of messy experiences and noisy input fuels "creativity". You don't think "I need to copy XYZ", but neither does the AI. You both just react.
In some regards our chaos is better, in others it is worse. But while the machinery of an LLM still does not even remotely approach a brain, we should not forget that we are nothing but more a cluster of small machines, assembled from roughly 750 MB worth of blueprint.
I wonder if this won't drive a resurgence of demand for live performances - as recording becomes more and more artificial, live performance will mean more. (Or maybe, as a live performer, I'm just wishful thinking here...)
This is not that. We're not talking about some inane mashup, but a wholesale digestion of every creative thing any person ever did by a monster computer cluster whose scale dwarfs imagination, which then promptly uses it to maximize "engagement" to gather eyeballs to feed them advertising. It's profoundly messed up.
It isn't. If a serial killer spent a week digging mass graves by hand, they don't get years taken off their sentence. You don't get points just for working hard or spending money, particularly when it cheapens or just appropriates other people's work.
> which then promptly uses it to maximize "engagement" to gather eyeballs to feed them advertising.
This is the real problem, right? People don't dislike generative AI, they dislike the attention economy. Yet I see more disgust towards AI than the company policies which suck. I don't understand why.
I think it is more that art, film and music have largely been replaced with complaining online about various subjects as the major form of entertainment in America.
Oh, haha, yeah. I guess I'm the opposite--I actually like AI more than the attention economy! At least one of them is not actively trying to drain my brainpower and skill set and get my to buy stuff and do stuff I wouldn't otherwise buy or do.
People also differentiate heavily on the basis of scale and profit. Artists are often fine with people sharing their posts and may even tolerate someone asking for permission to make printouts or whatever else for their circle of friends, but will expect some sort of royalty if you're asking to be able to sell prints of their artwork on a store.
Hell, even with viral videos it's relatively common that normal people can share away while entertainment companies and influencers are expected to pay for a license.
With memes it isn't clear exactly who made the first template, and the creation of them doesn't revolve around specific people in the same way, nor are they meaningfully tied to profits.
When creators post their content online to be shared, they do it with the focus being on reaching individuals, not for it to be sucked up by soulless companies to extract all value without the intention of giving back.
The conversation is quickly devolving into a vacuum of ignorance where things like royalties, fair use policies, revenue-sharing agreements, parodies, sampling, etc, have apparently never been thought about.
We're not talking about any of those things. We're talking about wholesale digestion of the entirety of human knowledge by automated means, which is now not just theoretically possible, but routine.
AI isn't influenced. It doesn't have restrictions. It doesn't have to work within confines. AI can always remember the word it wants to use. It always can hit the note it intends. And it can hit every note. Etc. It uses the corpus of training data and mashes it into a new form.
Stephen King won't be able to remember every word of every story he's ever read. And if he wants to make something "Lovecraftian", it'll be what Stephen King thinks is Lovecraftian. And there will be something to that. Some bit he believes is more or less important than other people And those bits are what makes Stephen King, Stephen King.
Everyone has had access to the same material King read. Access to the same tools he used to create. Everyone had the chance to effectively be Stephen King. But there is just one. Because there is some unique bit of observation or recall or combination of such things that is unique to King.
And from what I've seen so far, these LLMs can't do that. There is a missing element of pure imagination.
How am I doing that? I am claiming that LLMs lack imagination. They are incapable of creating out of whole cloth or interpretation.
Saying they cannot create based off of a vague suggestion is very much in line with that claim. I consider it a vital difference between Stephen King being inspired and LLMs mashing training inputs together.
