Fenway03

Awwwww.

An invisible watermark is never going to work. “De-glazing” training images is as easy as running it through a denoising upscaler. If someone really wanted to make a LORA of your art, Glaze and Nightshade are not going to stop them.

If you really want to protect your art from being used as positive training data, use a proper, obnoxious watermark, with your username/website, with “do not use” plastered everywhere. Then, at the very least, it’ll be used as a negative training image instead (telling the model “don’t imitate this”).

There is never a guarantee your art hasn’t been scraped and used to train a model. Training sets aren’t commonly public. Once you share your art online, you don’t know every person who has seen it, saved it, or drawn inspiration from it. Similarly, you can’t name every influence and inspiration that has affected your art.

I suggest that anti-AI art people get used to the fact that sharing art means letting go of the fear of being copied. Nothing is truly original. Artists have always copied each other, and now programmers copy artists.

Capitalists, meanwhile, are excited that they can pay less for “less labor”. Automation and technology is an excuse to undermine and cheapen human labor—if you work in the entertainment industry, it’s adapt AI, quicken your workflow, or lose your job because you’re less productive. This is not a new phenomenon.

You see, this quote of yours, ...

Artists have always copied each other, and now programmers copy artists.

… this is you being a capitalism bootlicker par excellence. And also, this is you being rather clueless about both art in general and the technology behind GenAI models in particular.

The process of one human artist "copying" or getting inspired by another involves (among other things) two key components: feeling and time.

The "feeling" component fuels this whole process, whereas the "time" component adds a sense of urgency and value.

An artist doesn't just randomly pick the color palette of another artist, the brushwork of a second, and the preferred motif of a third, then mashes it all together and ta-daah: new art!

Instead, when we draw inspiration from others while creating our own art, we "copy" those parts and styles of existing works that specifically speak to us, the parts that make us feel joy, sorrow, hope, despair, and everything in between. And the reason why we create our own art is because we want to share with the world how we feel. We're looking for a connection, for people who feel like us, for someone who gets us so we'll feel less alone.

A machine doesn't have any of those feelings. It doesn't understand the joy of devouring a strawberry sundae on a hot summer day, and it doesn't know the pain of seeing the light leave a loved one's eyes on a foggy November night. It doesn't need the company of another machine to feel whole. It simply doesn't feel.

And an AI art machine isn't constrained by time either. You could "freeze" its code at any particular moment, copy it to another set of hardware, and let it continue its work as if nothing happened. You could also copy this code to a hundred additional sets of hardware, and then you have a hundred machines performing the same work.

You can't do that with a human artist.

It takes time to learn a craft. It takes time to perfect it. It takes time to teach everything you know to those following in your footsteps so they can continue your work. And when you run out of time, you can't just take a snapshot of your brain, implant it into another body, and then live and create for another decade. That's why an original Warhol sells for $195 million, while a poster of the same piece printed in bulk sells for $19.50.

Feeling and time – that's what defines human art.

When "programmers copy artists", as you so ignorantly put it, none of the above process happens. In fact, those "programmers" don't really copy any artists at all. And they shouldn't even be called "programmers", because a key differentiator of GenAI is that it does not need a program in the traditional sense anymore. This is not how GenAI works.

In very, very simple terms, GenAI is based on a bunch of math and a shitton of hardware and data. When a model is being trained, it doesn't look at a Picasso and say "oh, that's an oddly interesting way to draw human faces!" Instead, the artwork gets analyzed pixel by pixel, first to detect very simple shapes and edges, then to detect certain combinations of shapes (e.g. those that look like an eye), then to detect even more complex shapes (e.g. two eyes, a nose, and a mouth that result in a face), and then even more complex shapes and relationships – until the model has detected and learned a specific combination of features that are often found in Picasso's art. And then you can let the model analyze an image it hasn't seen before, and it will tell you with a certain level of certainty whether or not that is a Picasso, too (i.e., whether the image contains all the patterns typically found in a Picasso). And that's also how you can let AI generate a picture that looks like a Picasso.

This whole process is just a lot of math (and computing power). There is no creativity. No feeling. No artist getting inspired by other artists while trying to express something meaningful about the world we live in.

Once the model has learned the typical patterns found in a Picasso, it can share this knowledge with another model – essentially within an instant –, and then the second model can identify and generate fake Picassos for you with the same level of certainty. It doesn't even need to learn anything anymore. [Again, this is a very simplified explanation.]

And this is what makes every capitalist's pulse soar: a "worker" that doesn't require costly training or years of experience, doesn't need sleep or breaks, won't get less productive with age, and could be cloned quite easily at low cost when the business needs to be scaled to meet growing demand.

"Art" generated by AI is a highly commodified business process. The only thing this kind of AI really creates is more money in a small number of capitalists' bank accounts. Because, as mentioned, it takes a shitton of hardware and infrastructure to train the foundational models. That's why the usual suspects (Google/Alphabet, Microsoft, etc.) dominate the field and can generate revenue every time some downstream developer uses the respective model for whatever AI application they want to build (from chatbots to image/video generators and so much more).

This also means (for now) that we should have a good laugh and walk away whenever someone argues along the lines of "but these new AI tools make art accessible to everyone; they remove the gatekeepers; now everybody can create whatever and whenever they want!" It could be argued that the opposite is true.

Now there are even more powerful gatekeepers. Filthy-rich millionaires and billionaires whose mass production tools make it even harder for artists to earn a living with their craft. And they have an army of bootlickers who run around spouting bullshit like "artists have always copied each other, and now programmers copy artists." Bootlickers who engage in victim-blaming and casually recommend that "you should unionize and demand that your labor is compensated fairly". (Gee, thanks, this has never been tried before and will surely be an easy solution.) Bootlickers who put the onus on those who can barely pay their bills instead of being mad at the rich assholes who exploit them. Bootlickers who pseudo-intelligently proclaim that "this is not a new phenomenon" while acting in a way that upholds the status quo.

The true grift here is not Glaze (regardless of its usefulness or lack thereof) – it's the "GenAI art is art" con that you have fallen for.

Since you're such a fan of AI, let's ask ChatGPT for advice:

In December 2023 they laid off a huge percentage of their translation teams, replacing them with ai and having the remaining members review the ai translations to make sure the translations are “acceptable” (Note how they use the world acceptable and not accurate)

Link to the tweet that informed me of this:

https://x.com/rahll/status/1744234385891594380?s=46&t=a5vK0RLlkgqk-CTqc0Gvvw

https://blog.duolingo.com/how-duolingo-experts-work-with-ai/

Duolingo has never been a platform that paid translators fairly. In fact, right from the start, they have shown zero respect for the translation industry and for the skills required to be a good translator. Their initial business model involved a crowdsourced translation service where they'd let language learners translate texts and then sell those translations to their clients (e.g., see this 2015 TechCrunch article).

Furthermore, the growth of Duolingo as a language-learning platform was only possible due to lots of naive language nerds who worked for free and helped create all those exercises through the company's Volunteer Contribution Program. That was a classic techbro asshole move on Duolingo's part: appeal to the good intentions of people with a passion for the subject/product, keep them motivated with lots of shallow talk about how their work will contribute to a brighter future where people have free access to knowledge and education, blablabla – and while all those well-meaning dummies worked for free, the people behind Duolingo were cashing in the big checks, getting investors involved, and planning the further commercialization of their product.

And make no mistake: The grand gestures they're now making, such as honoring those volunteers with fancy awards and VIP access to special events, and even the promised belated financial compensation to be distributed among all volunteers are just tiny drops in the bucket that won't hurt the company at all. And they won't undo all the exploitation that has been going on there for years. The partial switch to AI is just another non-suprising move following a long tradition of similarly profit-driven moves.

But they're not the only jerks out there doing this. Two more examples: Facebook (of course, eyeroll) and TED Talks. The latter's subtitles are also created by volunteers. And not only that: Many of those volunteers are actual translators (often with proper training and all) at the beginning of their career who unfortunately think that's a great way to build their portfolio and get some of that awesome exposure. But it is not. It's just a shitty way of helping all those rich tech companies get richer and further devaluing the translation profession. (If you happen to be a newbie translator reading this and looking to build your portfolio, do pro bono translations for people and organizations who really need your help!)

Obviously, this scheme is found in other fields as well, with people in creative industries being particularly vulnerable and gullible. Whenever someone promises your work will serve a greater good or provide you with career-boosting exposure, take a deep breath and then a close look at what kind of business or product you're about to support with your free labor. 9 times out of 10 you should be asking for real compensation.

And if you're a user of such products, 9 times out of 10 you should stop using them (and you should definitely stop paying for them).

But of course, life is complicated. Even if you should stop doing something, it's not always possible. Or at least not right away. (For example, I still use Facebook because of some non-public groups only found there.) But there's something you certainly can and must do: Pay more attention. Find out where your money is going, or, in the case of free services: Who will get your data, and how can they profit from them? Who's getting paid, and who's not getting paid? What's the history of the company or product? And do you really need it?

At least in some cases you will realize that you don't need the service or product at all.

AI in the form of large language models (LLMs) and generative algorithms used for tools like ChatGPT, Midjourney etc. increases this dilemma because it makes it even more difficult to find and use products whose creation didn't involve a lot of people getting exploited. And this will be the case for at least a while until the techbro hype has died down and people will learn to appreciate the value (well-paid) humans bring to a product or service. Lots of companies are currently trying to cut costs and corners by integrating these new AI models into their workflows. For some, it has already backfired (just ask the law firm Levidow, Levidow & Oberman about their little ChatGPT whoopsie); others will still learn this lesson the hard way. And of course, things will get reaaally fun when there's so much AI-generated content that the models will start ingesting too much of it, thereby poisoning themselves. Grab your popcorn, folks!

On a more serious note though: AI itself isn't the problem. It's an umbrella term that comprises a multitude of different methods and strategies, some of which are extremely useful (for example, in early-stage cancer detection). And there are many people, companies, and organizations that try to integrate AI into their workflows in a careful, cautious manner. You're already using lots of things in your daily life that wouldn't be possible without AI. Even your use of Tumblr is likely enabled by AI because fast internet requires smart routing of all that data traffic.

So from now on, when you look behind the scenes of how a tool or service gets provided, the mere fact that some AI is involved shouldn't be a disqualifier. You need to dig deeper. What kind of AI? What purpose is it used for? Does it actually help humans work smarter or does it force them to work harder?

There are problematic people on both ends of the spectrum: techbro bootlickers praising our AI overlords on one end, and uninformed luddites waving "boycott AI" signs on the other end. But a solution and a way out of this mess can only be found somewhere in the middle.