The Anamorphic Spiralcore of AI-Art
Goodlinks for 19/09/2023: Good AI-music still not here / Datasets as art / Mimetic AI-models of presidential candidates / Privacy in the age of AI / Harry Styles and the AI-demo-snippets and much more
The Anamorphic Spiralcore of AI-Art
In the last roundup i linked to a piece of AI-art called Spiral Town which since then made quite a splash in AI-art-circles. The Spiralcore-genre uses controlnet to influence the image synthesis with a pattern —in this case, a spiral and checker boards— and in combination with a prompt using architecture and landscapes, you get an effect which is called anamorphism in the artworld, "a distorted projection that requires the viewer to occupy a specific vantage point (...) to view a recognizable image".
This is an interesting technique for AI-art because there is no physical location that you, the viewer, could stand at to perceive this image which is only possible if you, the viewer, would stand at this particular location. Even if you'd create a CGI-anamorphism, you'd at least have the virtual location of the camera -- but not for image synthesis, which means the existence of this image alone is a nice paradoxical play with principles of AI-image making.
The original spiralcore prompter MrUgleh showed some contenders/experiment gens he made on the same day, and many people made variations of that: Here's Dreaming Tulpa with two sets of spiraltowns, Gavin Shapiro with a Summer cube, an exploring abandoned basements and a stable diffusion + control net animation test and singlezer0 animated some circular map thing.
Here's a Hugging Face Space by Angry Penguin to play with, and Ars Technica has a writeup about the phenomenon: Funky AI-generated spiraling medieval village captivates social media.
Datasets as Art
Pinned to my Tweetie-profile is a post saying that "Someone should put the 4.2GB Stable Diffusion model on a USB-Stick and tape it on a wall at Art Basel."
This tweet means that i consider Latent Space itself as a piece of art, an explorable ideaspace of a big chunk of art history which itself says a lot about the dissolution of boundaries of individual data points in a large interpolatable network and, by its open distribution, asks questions about the exploitative nature of this technology. I wish someone had actually done it.
Lila Shroff on Reboot is thinking in the same direction and writes about Datasets as Imagination. AI-artists since the late 2010s have created AI-art mostly by curating datasets and creating their own AI-models, not by prompting a huge image synthesis machine from a company. For instance, she writes about Anna Ridlers Myriad (Tulips), "an installation of thousands of hand-labeled photographs of tulips" she later used for the GAN-installation Mosaic Virus to talk about speculation bubbles in cryptocurrencies and capitalism, visualized with an algorithm reminiscences of the dutch tulip mania in the 17th century. This is arguably something else entirely and very distinct from, you know, "anime girl, cyberpunk outfit, very detailed, trending on artstation".
Shroff then goes on and writes about smaller, more specialised and more whimsical datasets as their own tools of imagination, which can capture more nuance and reject the idea of total representation of the billiontrillion parameter corporate image synthesis monsters.
The whole piece is a welcome alternative take on the well trodden paths of the AI-art debate and introduces a way to think about data spaces themselves as tools for creative expression.
Also, as it is the declared mission statement of this blog to keep things at least somewhat lighthearted, here's a data space worth considering for creative exploration from the Sex Tech Hackathon in 2016: Look at This Beautiful Algorithmic Visualization of Vibrators.
Persistence of Vision – Subverting civil infrastructure: Cool public art installation by Michael Candy about AI-enabled surveillance in cities -- a "network of spotlights disguised as security cameras. Each camera uses DeepStream vision software to detect and track humans as they pass through the laneway (...) The network is setup to alert neighbouring units of humans approaching. Allowing for the installation to ‘pass’ people between spotlights."
I’m a fake brand, in a fake world: The secrets behind designing a great fictional brand for TV and film: Interviews with fake brand designers who make stationary and fake media for movies. The props for Wes Andersons movies in the post are top notch. I just recently linked to thisthis video of Adam Savage visiting an old print shop producing fake media props for Hollywood.
Bill Willingham releases his beloved Fables-comic into the public domain because DC Comics suck. I'm very much over the whole "fairytales in modern settings"-thing that exactly this comic started, but this is a bold move. Respect.
HOTXTEA creates streetart with light, yarn and fences:
Mimetic AI-models of the 2024 presidential candidates are not a fun toy
Chat2024 tries to convince you that you can have "meaningful conversations with clones of the 2024 Presidential candidates" and i think this is exactly the dangerous stuff undermining a democratic process people have warned about. This website is promo for the digital cloning company Delphi with which "Creators, Coaches, & Experts (...) create a digital copy of themselves to talk with their fans 24/7/365 on any platform."
Now, I have zero problems with creators and coaches to create AI-clones of themselves because their fans and customers are adults and should know better than blowing their money into a mimetic AI-model, but i already get a bad feeling at the word "experts" and i think it's just ridiculous to provide AI-clones of politicians in a representative democracy to answer questions of the populace, and i think that this is dangerous to democracy because i'm sure if such a thing is deployed at scale, many people will believe that the AI-output of these models are somewhat averaged-but-authentic stances of the real politicians themselves.
For instance, a dimwit like orange clown may not look like the unarticulated, stupid ass motherfucker he really is, but an elaborate man who can formulate a coherent sentence. Or vice versa, an intelligent, decent, original and sometimes even funny politician may come around as bland, generic and shallow.
As these mimetic AI-models of real people have subtle effects on your psyche, mimetic AI-models of politicians will have subtle effects on your voting behavior. And because i like democracy a bunch, i am quite horrified here.
Daniel Dennett recently called counterfeit people the most dangerous artifact coming out of AI-technology, and counterfeit politicians pump the volume of this one up to eleven.
Actually good AI-Music is still not here
Stability released Stable Audio, a diffusion model for AI-music and sound. You can play with it here, and read about some technical background on TechCrunch and here's a short first guide on prompting Stable Audio.
I played with it for five minutes or so and while this may mark a jump forward for AI-music in general, it's output is far from being actually good music. I tried to get some punk and metal out of it but failed, the result being nothing but a muddy mess. The issue still seems to be a coherent song structure, which is detoriating into a weird chaos after a few seconds, and, as in previous music models, vocals are just gibberish (which makes vocals the weird hands of music synthesis). Not very surprisingly, techno and EDM seems to work just fine, likely because the structure of electronic music is much less complex and easily picked up by an algorithm.
Generative AI plugins for music production already are a thing, and i bet we'll see widespread adoption of AI-plugins in audio software like Traktor soon, providing on the fly generation of loops and what used to be called "DJ-Tools". If they can implement mapping for longer generated AI-tracks for use in sets in comming releases, this might be widely adapted by DJs very fast. Just like Jeff Mills is using a 909 in his live sets, he can generate loops and basic tracks on the fly by prompting the machine. Especially rave and techno and synthetic music are a match made in heaven.
Still, all generative music i've heard yet sounds super generic and the one thing you want from music is to catch you with surprise. For loops, samples and DJ-tools this is alright, but I don't think we'll hear a true actually good purely AI-generated song in a very long time, for the simple reason that music is much more information dense than images or video.
True coherent and purely generative AI-music has to solve for audio quality and complexity of song structure and the additional dimension of time, which increases the problem for AI to produce coherence many orders of magnitude. Image synthesis, by far the most advanced media synthesis AI-systems yet, only had to solve for image quality and complexity of composition, but not for a dimension of time. Equally, all good video synthesis as of yet is remarkable exactly for it's incoherence or seems to be heavily edited.
Music encodes much more information than (moving) images. We immediately can hear when something is off in music, where we are usually pretty generous with low information density in moving images, as every animator can tell you: look at anime and watch the actual frame rate of changing objects or the mostly static backgrounds. Sometimes faces or objects don't change for seconds and you just don't realize it, because it doesn't matter much. In audio and especially music, stuff like this is unforgivable. We may watch an animation in which objects change once per second maybe and not think about it -- but we never listen to a stuttering song.
In other words: The problem of audio information density in pure generative MusicAI seems far from solved, and Stabilities Audio Diffusion has not really solved this problem.
What you can get out of it is samples and techno loops and maybe some coherent but boring Ibiza house track, but not much else. However, not all is bad: To the delight of electronic music lovers worldwide, Paul Oakenfold may be out of a job very soon.
The Specter of AI-Generated 'Leaked Songs' Is Tearing the Harry Styles Fandom Apart: The future of music is hundreds of leaked demo-snippets by Harry Styles which may or may not be created by AI sold for hundreds of bucks to superfans. This is funny.
DJ Earworm mashes Deep Fake Drake, Empire of the Sun and Billy Ocean to sing Tracy Chapmans Fast Cars. Tracy Chapman and Billy Ocean are legends of music that even an abomination like Drake can’t diminish.
Some months ago i wrote about how AI will automatize mostly what David Graeber termed Bullshit Jobs: middle management, a lot of standard office work, writing elaborate sounding emails, summarizing excel sheets. Ethan Molick now did a study with a consulting company and the results are in: All consultants using AI were harder, better, faster, stronger. All you need to found a competetive consulting firm today is GPT, a suit and a shallow businessman smile. Looking forward to the coming automation bloodbath in the white collar bullshit job sector *evil grin*.
These Prisoners Are Training AI. I mean sure that's one way of adressing the AI labor issues, i guess.
"The French National Assembly has proposed a law to regulate the use of artificial intelligence (AI) in relation to copyright. The law would require AI software to obtain permission from authors or rights holders before using their works, and any resulting creations must be labelled as 'generated by AI' and include the names of the original authors. The law also proposes a tax on companies that exploit AI-generated works whose origins cannot be determined, with the proceeds going to (a) collective management organisation". //
I've written about how collection societies like IFPI or GEMA and copyright in general are not up to task to handle an interpolatable latent space, and this proposed law shows it. These demands, especially the naming of "original authors" in AI-output, can't ever be complied with and if this goes through, it's a death knell for generative AI in France. The copyright debates surrounding our beloved Large Language Warez gonna entertain us for a very long time, it seems.
Related: Pulitzer Prize winner and others sue OpenAI: "Pulitzer Prize winning US novelist Michael Chabon and several other writers are the latest to file a proposed class action accusing OpenAI of copyright infringement, alleging it pulled their work into the datasets used to train the models behind ChatGPT."
The right to be forgotten in the age of AI: Besides the ticking timebomb of copyright looming over generative AI, there's also the issue of privacy, as LLMs and Image Generators are trained on tons of private data, including medical records and biometric information. One way to tackle this problem for AI devs is Machine Unlearning, which are methods to delete information from a model without retraining from scratch. The new paper Learn to Unlearn: A Survey on Machine Unlearning looks at the state-of-the-art and finds that, while "machine unlearning can remove sensitive data or biases from a trained model without significantly impairing its performance", the tested techniques still need to improve to be compliant with privacy laws.
Here's a cool SDXL Vintage Magazine Style LoRA Available on CivitAI and Hugging-Face. I love these old pulp mags and if you want a real treat that is absolutely not synthetic, help yourself to the Flickr sets of James Vaughan where you can get lost for hours browsing nothing but old time vintage pulp mags and thousands of weird ass illustrations. You could train a million LoRAs on this flickr account alone. You're welcome.
There's multiple videos making the rounds of AI-translated videos with people's lips in synch to the synthetic voice in a foreign language. Here's one on Reddit, here's a guy on Youtube. I think i linked to the corresponding paper in the past but i can't find it, but anyways: These vids come out of an app called HeyGen which just introduced a new translation tool aimed at content creators. Pretty soon, all your beloved Youtube Influencers gonna speak all languages and thank god i'm not watching any of those.
I also bet that Netflix and any other movie production will soon provide movies in all languages without the need for dubs or subtitles. Not sure if this is progress in cinematic excellence, but i'm not a movie purist. I hate subtitles and prefer a good dub if i can't speak the language. While I do watch movies in their original language, when it comes to other foreign languages than english, i prefer a dub because i simply don't want to read while watching an audio-visual experience.
However, the quality of the deepfaked lipsync in those translation vids is far from being good enough for highdef movies, let alone 4k, so this tech has to improve a lot before Netflix can automatize dubbing and voice actors lose their jobs to AI.
Stephen Fry: AI copied my voice from Harry Potter audiobooks: "(This AI-model) could (...) have me read anything from a call to storm Parliament to hard porn, all without my knowledge and without my permission. And this, what you just heard, was done without my knowledge. So I heard about this, I sent it to my agents on both sides of the Atlantic, and they went ballistic—they had no idea such a thing was possible. (...) One thing we can all agree on: It’s a fxxxing weird time to be alive."
Ruby Thelot shines a light at the looming AI-flood: 10 Ways Human Writing Will Survive The Textpocalypse: "Once the machine has extinguished all the possible combinations of SEO-optimized articles, written and published, what will be left to write but earnest, profound, and true writing?"
The Rise of Tech according to Sandra Bullock movies: "Tech’s tendency to worm its way into intimate moments is not always such a wonderful thing."
Can ❤️s change minds? How social media influences public opinion and news circulation: While engagement numbers don't influence the opinions of casual users, socmed heavy users do change their opinions depending on perceived support (number of likes). Socmed heavy users are also more politically active. Go figure.
X still throttles links to Facebook, Instagram, and other competitors and The Markup has a tool where you can check various domains for Twitter-throttling. In other news, Elon Musk still has absolutely no balls whatsoever.
World Bank spent billions of dollars backing fossil fuels in 2022, study finds. Another reason to paint it black..
EU backs new air pollution limits, but not until 2035: "The EU has voted to clean its air by setting tough limits on pollutants in a move that experts say will save lives and money." // As you would expect, the car-industry backed conservative parties voted against saving lives and money.
The new game Solarpunk is one of those survival building games, but in a solar tech and post-climate change setting and it looks pretty good honestly if you're into those games.
Is air conditioning making cities hotter? Yes, it is.
California is suing Big Oil for deceiving the public and wants to make them pay "their fair share for: Recovery efforts from climate change-induced superstorms and wildfires; Protecting people from the health impacts of extreme heat; Managing dwindling water supplies in extreme drought; Fortifying infrastructure and homes against sea level rise and coastal and inland flooding." Good.
Kurzgesagt/In a nutshell on Boltzmann Brains: You're a Dream of the Universe (According to Science): "Absolutely everything you think about yourself and the universe could be an illusion. As far as you know, you are real and exist in a universe that was born 14 billion years ago and that gave rise to galaxies, stars, the Earth, and finally you. Except, maybe not."
Here’s the sources for this video on Boltzmann Brains.
Mathematician Steven Strogatz talks to neuroscientist Anil Seth about the Nature of Consciousness. I like the part where Seth compares his current work with the behavior of flocks of birds, which plays very nicely with recent comparisons of consciousness as emerging from the synchronized vibrations of matter. Ofcourse, Strogatz wrote the defining book about the mathematics of synchrony where he also has a chapter on memetics too, which in my view is the synchronization of more than one consciousnes. The podcast unfortunately doesn't go deeper into this, but it's still interesting throughout.
While we're at Anil Seth, here's an old video i just stumbled up with him talking about How We Build Perception from the Inside Out.
Collective intelligence may have shrunk our brains: "human societies got so cooperatively organized in the past 3,000 years that we began relying on what researchers call collective intelligence. 'It is the idea that a group of people is smarter than the smartest person in the group,' said James Traniello, a biology professor at Boston University and one of DeSilva’s co-authors. 'So basically, if you live in a group, you solve problems more rapidly, more efficiently and more accurately than what’s possible for any individual.' Traniello said the inspiration for applying this idea to why human brains may have shrunk came from 'ultrasocial' insects such as ants. Ants form highly cooperative societies in which division of labor has favored smaller-brained individuals due to an advanced level of social organization." //
I wonder if ultracooperative tools like digital networks will enhance this effect of shrinking brains, with additional evolutionary relaxation on the need for brain size coming from cognitive skills taken over by AI.
Decoding Morality: How Our Brains Differentiate Right from Wrong: "Neural evidence showed that liberals and conservatives have differential responses when judging moral foundations, reflecting their unique value emphases". //
I use to think of political preference as psychological differences, mainly. If you're on the right, your brain works different from someone on the left, which makes persuasion such a hard job. Convincing the "other side" of certain values or matters of importance to you requires empathy for their way of thinking, as hard as this may be. This study confirms this.
New Research on the Emotions of Highly Sensitive People: HSPs react much stronger to negative life situations and stress, but not on positive life situations: "increased sensitivity correlates only with greater emotional reactivity to negative events. Specifically, those with high sensory processing sensitivity experienced more negative emotions and lower self-esteem, self-efficacy, and life satisfaction in response to more subjectively intense negative events. They did not experience greater benefits from more subjectively intense positive events, such as larger increases in positive affect or life satisfaction.” //
I'm on the HSP-spectrum and this is bad news for me, but also aligns pretty well with my experience: Good stuff is good, but bad stuff is really really bad.
A whole bunch of cognitive scientists published a letter on Psyarxiv accusing Information Integration Theory of consciousness to be "pseudoscience" and magazines like Nature and Science to spread misinformation. IIT, in short, claims that any system capable of computing information to a certain degree of complexity can be conscious, including AI-systems and computers. It's a favorite among many cogscientists because it provides a way to calculate a level of consciousness which can be expressed in numbers, making it "hard science". Neuroscientist Anil Seth condemns the letter saying "The accusation of pseudoscience is serious & clickbaity & IMO the letter doesn't justify it" and Erik Hoes has a good takedown on it here: Ambitious theories of consciousness are not "scientific misinformation".
Panpsychist Phillip Goff in an interview on "cosmological fine-tuning, cosmopsychism (the idea the universe is a conscious mind) and whether the universe has a teleology, a purpose." //
Physics tells us that the universe at large is deterministic, the future is already existing and free will is an illusion — which makes your decisions and your free will ultimately much more important than you might already think —, and i think there is at least something to those points. But i also think that these things are outside of the scientific realm not only for the foreseeable future, but possibly forever, which means we are talking about the spiritual and belief here.
I come from a hard athetist background but have taken a somewhat softer stance on spirituality in the last few years, calling myself an agnostic these days, accepting that, yes, there are things we can't know and things that point into the direction that the universe as a whole might have a purpose. After all, i wrote about why we are neurons in a cosmic consciousness steering the universe along memetic fields towards the noosphere here on this blog. So i guess i at least somewhat agree with Goff.
Blind Test: 90s Techno / 303 Edition - Episode 30 // I got two of ten records right (Emanuel Top and System7/Plastikman) and i feel ashamed, because i owned a lot more of these.
Nicolas Cage is arguably on his way to become the greatest B-genre-movie actor of all times with his latest flicks being Dream Scenario, a horror-comedy playing with Freddy Kruger-vibes and Longlegs which just received a R-rating, in which he plays a serial killer with some occultism involved. Can't wait.
Wes Andersons The Wonderful Story of Henry Sugar: In which Wes Anderson continues his quest of hiding any human emotion behind stunningly accurate geometry of exquisite taste and absolute soullessness which are still fascinating to watch for exactly those reasons.
Here's the plot: "A beloved Roald Dahl story about a rich man who learns about a guru who can see without using his eyes and then sets out to master the skill in order to cheat at gambling."
Divinity is “set in an otherworldly human existence, scientist Sterling Pierce (Scott Bakula) dedicated his life to the quest for immortality, slowly creating the building blocks of a groundbreaking serum named Divinity. Jaxxon Pierce (Stephen Dorff), his son, now controls and manufactures his father’s once-benevolent dream. Society on this barren planet has been entirely perverted by the supremacy of the drug, whose true origins are shrouded in mystery. Two mysterious brothers (Moises Arias and Jason Genao) arrive with a plan to abduct the mogul, and with the help of a seductive woman named Nikita (Karrueche Tran), they will be set on a path hurtling toward true immortality.“
More trailers worth watching: The Fall of the House of Usher, Pet Sematary: Bloodlines, Aquaman and the Lost Kingdom, Nowhere, The Creator Final Trailer, Monsters of California, a Blink-182 Alien Conspiracy Film, The End we start from, The Elderly
Researchers found a way to write into water, here's the paper: "The study here demonstrates the writing of lines, letters, and complex patterns in water by assembling lines of colloidal particles. Unlike established techniques for underwater writing on solid substrates, these lines are fully reconfigurable and do not require any fixation onto the substrate. Exploiting gravity, an ion-exchange bead (pen) is rolled across a layer of sedimented colloidal particles (ink). The pen evokes a hydrodynamic flow collecting ink-particles into a durable, high-contrast line along its trajectory. Deliberate substrate-tilting sequences facilitate pen-steering and thus drawing and writing."
I'm not an expert in waterwriting, but i think there's a much much simpler technique to write in water: Use a metal pen cooled way down below the freezing point and create ice writing. You just have to find a way to not make the cooled down pen gets stuck. But sure, nanotechnology *eyeroll*.
Reanimated spiders and smart toilets triumph at Ig Nobel prizes: The Ig Nobel Prize from Harvard University "honors achievements that make people laugh, then think" and is given to researchers for the bizarre, funny and weirdest scientific papers and on my old blog, i wrote about them every year. This time they awarded a paper on robotics in which they used dead spiders as a sort-of-exoskeletton for robots, a study on the weird linguistic sensation you get when you repeat a word word word word word word word word word so often that i looses its meaning, for "measuring the extent to which ocean-water mixing is affected by the sexual activity of anchovies" and "for studying the mental activities of people who are expert at speaking backward". I love the Igs. The full list of winners is here, a video of the prize ceremony is here if you like to watch academics with silly hats celebrating weird science.
"Exactly 10 years ago (Martin Kleppe) revealed 'WORLD'", "a mesmerizing 3D globe encapsulated in its own source code! This quine is crafted in just 1024 bytes of pure HTML + JavaScript, with no external dependencies." // I remember blogging this back then.
Welcome to RamenHaus, home of rotating ramen. More rotating food please.
A clock where the time is made of news headlines
21st Century Roguelike Pac-Man
Via Webcurios, another one of those CGI-artist-battles, this time it’s the Top 100 Boss Fight 3D Montage.
Rich people fix their teeth so hard it becomes creepy: Celebrity teeth look the same thanks to veneers, and TikTok has thoughts. //
I'm what most people would call poor, and i live with poor people, and i can tell you that none of us have this particular kind of problem and our unperfect, uncreepy grin goes out to all the rich and the beautiful.
Btw, find below many options to support this blog.