Discover more from GOOD INTERNET
I don't buy into the flood because Synthetic Taylor Swifts lack drama.
Plus: the state of the world 2023 / rickroll ckpt / these artists do not exist / nietzsche at a rave / family portraits with analog synthesizers / the physics of baking good pizza / and much more
One reason why i don't believe in the AI-content flood is because it's not engaging.
Bots are an overblown problem and their impact is negligible. It just isn't very engaging to talk to a machine, and socializing on the web only rarely works by stumbling over a random account you like enough to follow, bot or not. You follow people you know from somewhere or you have seen on other platforms in other contexts maybe and who post stuff that is of interest to you, and then you follow the people who engage with them — you follow because you trust, at least to a certain extend. The percentage of bots i have been following over the years is near zero, except for some automated art bots. I believe i only rarely talked to a bot without knowing, if ever. As with spam and desinformation, the possible volume of production says nothing about impact.
The possible volume of AI spam is an endless amount of dick enlargement mails. AI already is deployed in the professionally shady spam business on a massive scale, the bulk of spam already was algorithmically produced for years anyways — and yet, i haven't read a spam mail in years, not to speak engaging with it.
I don't even think that we need something like AI steganography which is in development everywhere, because this stuff is not the way we share. We don't just randomly see something online and share it. We share because we like or hate some piece of content and (somewhat) trust the judgement of our online-peers. Studies are clear on this: Its the (perceived) competence authority of our peers that makes us share stuff, not just random stumbling on stuff. It’s about trust.
In other words: For me, its much more likely to share a piece by Stefan Rahmsdorf on climate change because it was shared by him or an account i know to share reliably good stuff, than a piece by a random person on Twitter. And for a rightwinger, its much more likely to share a piece by Bjørn Lomborg shared by a reliable source for rightwing sources, but they too don't really share from random accounts or Twitter bots. (Conservatives actually do fall more for fake news and bots, but i'm not sure the total number is really significant.)
Another angle to view this is the flood of millions and millions of synthetically images coming from all the image generators. Although there is the occasionally really artful image synthesis, and some of the stuff just blows your mind on first sight, it is remarkable how impactless the stuff really is, and how much the bulk of these synthetic images shows always the same fantasy warrior elve princess, which was uninspired and uninteresting derivative imaginary even before the advent of image synthesis, even if it shows 5 fingers on an anatomically correct hand.
What I’m saying is that for image synthesis, too, 99% of the stuff is just meaningless fluff that never will interest anybody and leave zero impact, never to be seen outside of a Discord server.
Or take the 100.000 Synth-Taylor Swift albums that surely will “flood the market” coming from black market CKPT-files that just churn out endless variations in the style of the pop singer and make her job “obsolete”. Really? Will it? What are you buying when you buy a Taylor Swift album? Are you really buying a data storage containing a bunch of audiofiles? Or are you buying an artistic artifact connected to a human story which can’t be synthesized? Those 100.000 Taylor Swift albums surely will exist, i have zero doubt about that. I have also zero doubt that i will never listen to them, outside from my interest in algorithmic aesthetics. Because those synthetically generated Taylor Swifts lack drama.
The Dadabots are an art project i wrote about years ago and they generate endless livestreams of any genre in music, and as an artproject this works because it speaks about algorithmic art itself. But as a piece of music you engage with, that touches you or makes you go wild in a moshpit? When was the last time you decided to sit down and listen to a robot, not because it was a gimmick, or because everybody talked about AI suddenly, or because you wanted to check out how robots sound, but because it was good music you would enjoy and which speaks to you, as a person?
I write about algorithmic art and music since forever and i follow every development in the art/tech-spaces and its intersections and i listened to exactly zero synthetically produced pieces of AI-music. Because it has no impact, and the stuff is not engaging. The same will hold for AI-content of all kinds.
So, yeah, there is a flood of bots and soon, there is a flood of synthetic AI content, but it's impact is irrelevant, because this flood breaks on shores we rarely visit anyways.
So no, i don't think that AI content will be a big problem for the general public, just as PR-speaking bullshit content is not a big problem and is largely ignored by the general public.
It's still people that rile us up, make us run up the walls, make us look in awe or feel cozy and loved. I don't think this will change with automated culture-mimicing expression, at least not in the foreseeable future.
A rickroll ckpt. the future gonna be wild. "This model is trained on 130 images [from Rick Astleys 'Never gonna give you up' musicvideo], 1200 steps UNet and 400 steps text_encoder."
Riffusion, a diffusion model for music. The "beautiful 20-step interpolation from typing to jazz" is especially juicy.
Stable diffusion can now generate audio transitions that are smoother than any human transition. Mapping tracks on Traktor will look very different in a few years, but transitions for DJs are more than fading in/out. When SD can cut and throw a backspin and go full Mixmaster Mike, then we talk.
A Stable Diffusion dreambooth that generates AI-Art-Protest images. Sometimes even gets typography right :)
“Introducing Bird SQL, a Twitter search interface that is powered by Perplexity’s structured search engine. It uses OpenAI Codex to translate natural language into SQL, giving everyone the ability to navigate large datasets like Twitter.”
“In order to protest AI image generators stealing artists work to train AI models, the artists are deliberately generating AI art based on the IP of corporations that are most sensitive to protecting it.”
That’s glasses: Record-high myopia solved by an alliance of experts: -108.00 D
Joshua Citarella published his eBook on "Politigram & the Post-left" for free on Substack. I blogged about this back then on my old blog when it was new, and it’s still as relevant as it was then if you’re interested in hardcore niche political radical subcultures on the web.
Me on Twitter: “We should consider every day lost on which we have not danced at least once”. (Friedrich Nietzsche) —> “Nietzsche dancing at a rave party”. (Midjourney)
Walking stupid is a form of exercise i can get behind: Quantifying the benefits of inefficient walking: Monty Python inspired laboratory based experimental study
Zach Lieberman, atlas of blobs: “For this project, Atlas of Blobs, I asked ten artists, designers, researchers, and visual thinkers to pick one of the blob forms I’ve made and write a text to name and describe it.”
If you think you’re bad at tweeting, think again: Worst Tweets 2022