The Colors of Her Coat
astralcodexten.com
astralcodexten.com
I do believe AI is a legitimate tool in a human artist's toolbox. I think it's one that's rather difficult to use well, but that isn't disqualifying per se.
That's true, normally you multiply them.
Presumably if you're OpenAI and choosing showing that prompt as an example, it's because you have an appreciation for the artwork that created that style. Shouldn't you show those creators the respect of asking for permission? Not as a legal obligation, but a moral one, out of the very sense of appreciation for the source material that makes AI-Ghibli-homage artwork feel charming?
Artists don't seem happy with the pipeline of companies digesting their work, usually without permission, into ML models that are used to generate thousands of imitation works. Perhaps it is the issue of commoditizing someone's painstakingly developed individual style, or simply of competing with the artists themselves with realistic commissions that can be generated rapidly at almost zero cost.
When people Ghibli-fy their family photos, they get a rush of warmth that isn't from the aesthetics alone (otherwise any equally-good art style would do), but instead specifically comes from a place of admiration for films that they love, and how heartwarming it feels to be able to see yourself through that lens.
The very reason these memes create good feelings is because of their connection to Studio Ghibli itself, and for a tech company to just ride that without even so much as asking permission transforms it (in my opinion) from an homage to a pillage. It doesn't make it illegal, but to me, it hollows out the warmth and positivity that ought to exist in these images, because it spits on the creator rather than honouring them.
You can also print the Mona Lisa on toilet paper, but that's a weird way to express one's love for da Vinci. And so part of what I'm saying is, this is not merely a question of abundance.
Compare with Studio Ponoc, composed of ex-Ghibli employees who also took the same unmistakable Ghibli art style with them for their first feature film. They talked with Miyazaki and got his blessing[1]. That may not have been a legal duty, but it gave their work a certain moral legitimacy, transforming it from imitation to an inheritance.
You don't have to look backwards very far to a time when artists Simpsons-fied or Ghibli-fied avatars of others or themselves, and they certainly ask no permission, at all, ever.
> Compare with Studio Ponoc, composed of ex-Ghibli employees who also took the same unmistakable Ghibli art style with them for their first feature film. They talked with Miyazaki and got his blessing[1].
Yeah it's a different ballgame when you're a peer in the animation industry.
Claude? Claude as in the LLM Claude? You couldn’t find good numbers and then choose to source an LLM for this information? My god, if you can’t find certain information and an AI magically can it could very very well be that it’s just making it up.
A Shepard tone can produce the illusion of a sound that continually goes up in pitch, yet never goes beyond the range of human hearing:
https://en.wikipedia.org/wiki/Shepard_tone
Maybe there is some intervention that can produce a hedonic illusion along the same lines: rising joy every day as if you lived your best experiences for the first time, joy even if you are not socially elevated over others, joy that cannot be diminished by comparison or repetition. Since it's logically impossible for everyone to be The Cleverest, The Most Popular, or The Person With the Most Attractive Spouse, (much less to reach those pinnacles for the first time every day), improvements in quantifiable material conditions will not durably quell dissatisfaction at the population level. No more than better eyeglasses will end optical illusions. We need material improvements and we need perceptual counter-illusions against our natural illusion that things never really get better.
Roughly I would explain the way this works as: humans have actual phenomenological experiences as bodies in the physical world, which acts as the ground-truth for our subjective experiences.
Then, on top of this ground truth, accrue many layers of additional information—ideas, prejudices, labels, etc. LLMs, when they parrot human speech or artistic mediums, are able to reproduce many of the patterns in these higher layers. Fine.
We are especially attuned (as kids, say) to the physical presence of other humans, and to other living creatures and the natural world. And—if William James is to be believed—the experience of a sunset, say, when unobscured by these layers of ideas, is a nearly divine thing—seemingly alive in itself, or a gift bestowed by some will outside ourselves.
An artist, then who is very attuned with the subjective "ground truth", and capable in their medium, is able to make art which represents that ground truth such that our human nature recognizes as unmistakably human. This is not just a blob of information which accurately references the human experience, but is furthermore crafted just so that it strikes us as of the ground-truth-world itself, subverting or transcending those layers of ideas that obscure the fullness of feeling, by being perfectly tuned to human nature (by their own human nature).
Something like that, anyway. Perhaps an LLM could copy this too. That LLMs so far have failed to, though, is telling—every single LLM output (and many human outputs) read as cliche, kitsch, or mimicry.
Even if we assume an LLM could capture the latent "ground truth" in human art, there is additionally the element of "being in the room with the art". A painting simply does not equal the "information in the painting"—it is a physical object, itself part of the same world our bodies occupy, and this is part of its essence. Great art, we might say, is that which strikes our own human nature as a recognizable part of the "ground truth of reality", as opposed to a mere arrangement of information.
Generally this is most easily seen with live music. I've only ever felt the "ground truth" I'm describing around classical music, myself, because that's what I know, but I've seen enough to know that the thing all these AI art squabbles are about ain't it.
https://en.m.wikipedia.org/wiki/The_Work_of_Art_in_the_Age_o...
That's pretty much the reason why "salvation" is not merely synonymous with "a permit to enter Heaven" (even though this part seems to be often omitted in the teachings): you still need to untwist your soul and realign it with the Heaven's nature.
It's not about making something that used to be difficult and scarce (obtaining lapis lazuli for creating the ultramarine pigment) into something that is readily available (synthetic ultramarine), but about saturation and cheaping of semantic meaning.
Studio Ghibli's movies are not valuable because they are scarce, they are valuable because there's meaning and joy in them that is not for instant gratification or cheap meme-making. Flooding the internet discourse with Ghibli memes is not "making the scarce more available", it's producing knock-offs that have the potential to saturate our collective minds and make them "inured" against the real Ghibli.
It's not about being a luddite or rejecting technological progress, it's about not wishing to see our culture destroyed with noise that drowns everything else. And yes, pop culture IS culture.
By the way, it's not about Ghibli, TFA-within-TFA also mentions Nabokov, just in case someone goes "I didn't like Ghibli to begin with, so what?".
I have the upmost respect for Bill Waterson for not merchandizing Calvin and Hobbes and the cultural memory of the work is better for it imo.
I feel like this claim rests upon several unstated claims that don't seem rigorously proven. What does it take for an art style to be "saturated" in culture? What would it mean for a brain to be "vaccinated" against it?
Hoel presents the case of semantic satiation, the idea that the repeated use of a word leads to a loss of meaning and a preoccupation with the syntax and structure of the word itself, as an analogue. But semantic satiation is a very short term effect. Having once engaged in semantic satiation does not permanently make words lose their meanings.
Similarly, I am really skeptical this short term fad of using OpenAI's current image model to replicate Ghibli's work will permanently cheapen the semantic meaning of Howl's Moving Castle or Spirited Away. Hoel quotes a twitter user that apparently feels this way, but I really think people are being hyperbolic about a short term experience. Having not been on twitter much the last week, I've only been dimly aware of the phenomenon and I would guess that it won't even make it to May as a mass phenomenon, though I'm sure a handful of people will remain obsessed with it. I just don't see how that could ruin a great work of art either for me or for the culture at large.
I think he has a particularly vile audience now and that's influenced him and his writing.
He is very explicitly anti-woke and this is stated directly in many articles.
Wrong.
> He is very explicitly anti-woke and this is stated directly in many articles.
He is quasi anti-woke to the extent "wokeism" becomes a self-defeating problem. He isn't "anti-woke all the time".
If you want to talk about him being explicit, then take stock of what is explicitly said.
Here is one of many examples. He talks all the time about how he hates wokeness and woke people. You can still vote democrat and be anti-woke in a way that's annoying to read.
Lol you realize I have eyes right? Did you even read this?
"For me, seeing actual injustices against minorities makes me more woke, and seeing woke people be stupid and unnecessarily combative makes me less woke." -- yeah, really oozes hate.
It's clear what this is: you find any hint of criticism for this pet issue heretical.
Well sure, it’s possible. But the dominant market & cultural forces win out in the long run. There are modern oil painters doing art that is technically better than “the masters”, but nobody knows their names because they’re orders of magnitude short of critical mass.
Always strange to see an author (I think Unsong was self-published? I don’t recall) act so callous towards the copyright of another artist. Perhaps it’s the change in media that makes it feel okay to him.
It's not copyright. Specific works of art are copyrightable, but a style is not. We probably do not want to extend copyright protection to style because its boundaries are fuzzy and because art advances through incremental improvements.
However, we give social value to originality. A human artist who could just reproduce the Ghibli style is 'lesser' than one who creates their own.
This speaks to the tension described by the author. On one hand, we deem that human creativity is worthy of status and that copied styles are 'imitation' or 'knock-off' goods. On the other hand, as consumers we gravitate to the familiar, and success breeds just those cheap imitators.
I very much disagree with this. Certainly no gen-AI model currently does this, but you could reinforcement-learn your way to Ghibli style. In language modelling, that's the entire point of the RLHF step.
As long as the style can be described or even recognized by humans, you can nudge your way towards it with multiple-choice sampling.
> such a tool is producing derivative works based on original work (let’s say, by Miyazaki or Studio Ghibli) at scale and for profit, then I have difficulty seeing how copyright law is not involved.
How is this different than a fairground caricaturist that draws subjects in the Simpsons style? They "ingested" lots of Groening-style media, and now they're creating works that are derivative of that style, at scale and for profit.
Learning from a work and being inspired by a work do not create 'derivative works' for the purposes of copyright. The boundary between that and an infringing quotation is fuzzy (note 'sampling' cases in music), but a style itself needs to be on the non-infringing side of things if art is to progress at all, even by humans.
Note that I'm leaving aside the issue of data-gathering; if OpenAI was downloading pirated copies of the Ghibli corpus then that act itself was copyright infringement, independently of the AI training and output.
> if OpenAI was downloading pirated copies of the Ghibli corpus then that act itself was copyright infringement, independently of the AI training and output.
On the contrary, having legally bought the works is irrelevant. You can buy a Disney movie and be liable if you screen it at a local mall you own.
The legal debate in the United States, where most of the relevant models are being trained, revolves around whether or not the use of training data is fair use or not.
That’s to say, it may be a legally permitted form of copyright violation, or it may be a legally prohibited form of copyright violation.
Either way, it is legally an issue of copyright. On a secondary level we can talk about trade dress and moral rights, but that simply muddies the water. The legal discussion centered on copyright is concrete and unavoidable.
This is all orthogonal to moral/ethical/societal concerns.
That's arguably not really accurate, since statutory fair use itself (and this is why it is written in a less straightforward fashion than most of the rest of copyright law) is a direct statutory codification of what the Supreme Court found to be a Constitutional limit (based on the First Amendment) on the copyright power.
Fair use is not a “legally permitted form of copyright violation”, it is the space where the federal government has no power under the Constitution to create exclusivity as part of copyright.
Where exactly is the copyright violation supposed to be occurring? Is it distinct from the 'copyright violation' that happens in the brain any time someone learns what a Ghibli frame looks like? What if I study Ghibli in an art class and write notes to myself about the art style, to jog my memory later?
If the problem is that model training is considered a weird kind of copying, what's the legal line between that (allegedly infringing) copy and taking a JPEG and re-encoding it as a PNG?
> *Chesterton’s answer to the semantic apocalypse is to will yourself out of it.* If you can’t enjoy My Neighbor Totoro after seeing too many Ghiblified photos, *that’s a skill issue.* Keep watching sunsets until each one becomes as beautiful as the first (the secret is that the innumerable company of the heavenly host sings in a slightly different key each time).
> I support Erik Hoel’s crusade to chart some society-level solution to the semantic apocalypse problem. You’re not allowed to say “skill issue” to society-level problems, because some people won’t have the skill; that’s why they invented the word “systemic”. *But your personal relationship to the meaning in your life is not a society-level problem.* While Erik Hoel works on the systemic issue, you should be thinking of your own individual soul.
> *If you insist that anything too common, anything come by too cheaply, must be boring, then all the wonders of the Singularity cannot save you.* You will grow weary of green wine and sick of crimson seas. But if you can bring yourself to really pay attention, to see old things for the first time, then you can combine the limitless variety of modernity with the awe of a peasant seeing an ultramarine mural - or the delight of a 2025er Ghiblifying photos for the first time.
Conditional on an AI actually being able to accelerate progress in these areas ( the likelihood of which the author does not even try to guess), the part of "helping other people" not surely remaining a true source of meaning still holds.
2. At a high enough level, isn't helping other people the most valid possible source of meaning? Finding beauty and sharing it? If AI can somehow impossibly replace humans ability to ever help another human then we are done. No meaning is possible. Nothing matters.
The article writes like this is an inevitability and I think that's weird and wrong.
I mean, even if an AI could one day actually do that, they would sell the service only for a price that makes you poor for the rest of your life.