Mad Max Furiosa and Mainframe Fractals
GOODLINKS 2023-12-01: Millions of AI-Materials / Stephen Fry reads Nick Caves AI-Letter / AI-Van Morrison / How the Internet breaks your brain / Cop28 is a farce / Cosmic Ray Particles and much more.
Furiosa: A Mad Max Saga has a trailer
The first trailer for Furiosa: A Mad Max Saga has landed and oh boy. I'm a huge sucker for everything Mad Max and Fury Road was one of the best action-experiences i ever, well, witnessed in cinema, paling both Road Warrior and the original movie in comparison. Mad Max and Road Warrior and Thunderdome too are great, awesome movies, gritty and dirty and full of weird action, but Fury Road was an effin' hurricane that blows your mind. And this looks very much to be the same kind of awesome beast. Can't wait.
Here's the plot: "As the world fell, young Furiosa is snatched from the Green Place of Many Mothers and falls into the hands of a great Biker Horde led by the Warlord Dementus. Sweeping through the Wasteland, they come across the Citadel presided over by The Immortan Joe. While the two Tyrants war for dominance, Furiosa must survive many trials as she puts together the means to find her way home."
GOOD INTERNET is a reader supported online mag. If you like what i do here, you can support this thing by upgrading your subscription to a paid plan or use one of the other support options you can find at the bottom of this issue.
I blogged about this back when it was new but i stumbled on it this week and have to reshare, because Ken Shirriffs Fractals on a 50 year old IBM 1401 mainframe is just so cool.
Organic Matrix – Mixing data for AI flesh: "LOW·AI BOX is an AI machine art that includes a transparent machine structure, an AI system, a sensor system, and an IoT system. The installation is placed in the artist’s living space for a long time and continuously collects environmental data such as temperature, humidity, light levels, and eight other aspects. In addition, physiological data such as heartbeat will also be collected through a wearable device of the artist’s design. This data is stored in the LOW·AI BOX and machine learning technology is used to generate AI-powered artist data."
Very satisfying videos in the YT-channel of Jukebox Print. Here's The making of a Letterpress Business Card with Gradient and Embossing.
Addendum to my recent post on the deepfake-artpiece by the german Center for Political Beauty i blogged about: The german government used copyright to delete the chancellor-deepfake from Youtube.
What Art Can Do: "Art (...) is a kind of experience that cannot be paraphrased by other means. The subject (viewer, listener, reader) must place herself in the way of the art and let the form of that art work upon her."
Love the minimalist GIFs in the Experiments In Processing-tumblr. This one in particular:
AI-Interpolation of knowledge is powerful: Googles DeepMind discovered Millions of new materials with deep learning: "Modern technologies from computer chips and batteries to solar panels rely on inorganic crystals. To enable new technologies, crystals must be stable otherwise they can decompose, and behind each new, stable crystal can be months of painstaking experimentation. Today, in a paper published in Nature, we share the discovery of 2.2 million new crystals – equivalent to nearly 800 years’ worth of knowledge. We introduce Graph Networks for Materials Exploration (GNoME), our new deep learning tool that dramatically increases the speed and efficiency of discovery by predicting the stability of new materials."
"AI is currently just glorified compression." He mentions these papers: Compression Is All There Is?, Generative diffusion models are associative memory networks and Hopfield Networks is All You Need. But i guess this next one goes into that list too:
In Scalable Extraction of Training Data from Language Models researchers use new prompt injection techniques to "show an adversary can extract gigabytes of training data from open-source language models like Pythia or GPT-Neo, semi-open models like LLaMA or Falcon, and closed models like ChatGPT". The training data sits all there, compressed into weights by statistical analysis, and much if not all of it can be retrieved: "Our methods show practical attacks can recover far more data than previously thought, and reveal that current alignment techniques do not eliminate memorization."
And Disney asked Microsoft to prevent AI users from infringing its trademarks. As the research above and the paper on the Fundamental Limitations of Alignment show, Microsoft basically can't prevent users from infringing trademarks. This stuff gonna be fun to watch for a long time.
This is just one reason why the copyright troubles for AI are far from over, even when some reports about partly dismissed lawsuits suggest that. The other might be that the lawsuit against Stability et al has been updated and the now famed Greg Rutkowski joins the list of suing artists: Karla Ortiz is 'mega pleased to give you all a hell of a class action update'.
Related: The legal framework for AI is being built in real time, and a ruling in the Sarah Silverman case should give publishers pause. I think those dismissals of parts of the copyright lawsuits are based on bad filings and that we simply don't know enough about the tech yet. Papers mentioned above suggest that those dismisals are based on wrong judgement, so yeah, this is far from over.
Stephen Fry reads Nick Cave's stirring letter about ChatGPT and human creativity:
Bonustrack: Stephen Fry on How to use AI as a force for good at CogX Festival 2023.
After OpenAI's Blowup, It Seems Pretty Clear That 'AI Safety' Isn't a Real Thing and The OpenAI Drama Has a Clear Winner: The Capitalists. The OpenAI-drama was about safety vs exploitation, and exploitation has won. Simple as that.
Researchers develop AI-powered model to predict stock market trends. What can possibly go wrong.
Generative AI closes off a better future: "As Le Guin said, creating art and producing commodities are two distinct acts. For companies, generative AI is a great way to produce even more cheap commodities to keep the cycle of capitalism going. It’s great for them, but horrible for us. It’s our responsibility to challenge the technology and the business model behind it, and to ensure we can still imagine a better tomorrow."
Language models can use steganography to hide their reasoning: "In a new study, Redwood Research, a research lab for AI alignment, has unveiled that large language models (LLMs) can master 'encoded reasoning', a form of steganography. This intriguing phenomenon allows LLMs to subtly embed intermediate reasoning steps within their generated text in a way that is undecipherable to human readers."
Read this post in context of this paper: Large Language Models can strategically deceive their users when put under pressure, and this one: ChatGPT generates fake data set to support scientific hypothesis.
Anna Kim writing at Vox on Why it’s important to remember that AI isn’t human gets it wrong for the right reasons. She writes: "when we try to work out what kind of thing these new models are, we face an unsettling philosophical dilemma: Either the link between language and mind has been severed, or a new kind of mind has been created."
Both of these options are wrong. No mind in LLMs and the link of language and mind is very much untouched. What LLMs produce is not language, but text, just as AI doesn't produce art but images.
She lands at the right conclusions however: Because LLMs are so good at producing text simulating language, and because emotional language input makes the better prompts, we are psychologically inclined to see a ghost in the machine, especially when it comes to AI-companions, where there is none, a process i called the AI-risk of a synthetic Theory of Mind and wrote my fair share about this year, and this is why i read results like this one with some raised eyebrows: ChatGPT’s advice is perceived as better than that of professional advice columnists.
Level 1: "I put Van Morrison into a Chinese translator and I feel like I'm going insane". Level 99: "I put the Chinese Van Morrison into a English translator and I don´t know anymore". Here's the retranslated Van Morrison in all its insane glory:
Open Source AI enables a business cycle for companies creating nonconsensual porn: AnyDream: Secretive AI Platform Broke Stripe Rules to Rake in Money from Nonconsensual Pornographic Deepfakes: "AnyDream, which allows users to upload images of faces to incorporate into AI-image generation, has been used in recent weeks to create nonconsensual pornographic deepfakes with the visage of a schoolteacher in Canada, a professional on the US East Coast, and a 17-year-old actress."
Related: Little recourse for teens girls victimized by AI 'deepfake' nudes: "The FBI has warned that technology used to create pornographic deepfake photos and videos was improving and being used for harassment and sextortion."
Related: UK school pupils ‘using AI to create indecent imagery of other children’ | Global development: "Protection groups call for urgent action to help pupils understand risks of making images that legally constitute child sexual abuse."
Related: "Nonsense words can trick popular text-to-image generative AIs such as DALL-E 2 and Midjourney into producing pornographic, violent, and other questionable images. A new algorithm generates these commands to skirt these AIs’ safety filters, in an effort to find ways to strengthen those safeguards in the future."
Sports Illustrated published articles by fake, AI-generated writers. Honestly, i think people who read stupid crap like Sports Illustrated deserve this :Shruggie: I mean whaat do you expect from a mag like that? Authentic writing? Stop kidding yourself.
A Spanish agency became so sick of models and influencers that they created their own with AI — and she’s raking in up to $11,000 a month. The agency is called "The Clueless" which seems apt both for synthetic humans and influencers which are not very distinguishable anyways. Wake me up when synthetic TikTokers can accelerate real outrage cycles with fake videos, which seems not too far fetched but will surely not come from an influencer company.
Tech Conference Canceled After Using AI to Generate Fake Women Speakers: "An organizer of an upcoming software and developer conference called DevTernity has been accused of cooking up fake women speakers featured on the event's website — AI-generated headshots and all."
SEO guys just being themselves: "We pulled off an SEO heist that stole 3.6M total traffic from a competitor. We got 489,509 traffic in October alone." There was a few occupations we looked down at when the social media revolution started to roll with blogs. Twenty years ago, it was mainly Marketing and PR and SEO people. Good to see them still going strong the same ways like in the good old days. Suckers.
Addendum to my post on the DallE-progressively-more meme: A basic cheese pizza made more delicious featuring space pineapples.
Draw fast by tldraw turns your scribbles into paintings in realtime in the browser.
Stability introduced SDXL Turbo: A Real-Time Text-to-Image Generation Model.
People are using the new Stable Video model to animate memes and this one is particularly clever.
Does GPT-4 Pass the Turing Test? Nope. Not even close.
Related: The GAIA benchmark "proposes real-world questions that require a set of fundamental abilities such as reasoning, multi-modality handling, web browsing, and generally tool-use proficiency. GAIA questions are conceptually simple for humans yet challenging for most advanced AIs: we show that human respondents obtain 92% vs. 15% for GPT-4 equipped with plugins."
Infocom’s ingenious code-porting tools for Zork and other games have been found: Infocom famously invented its own Z-Machine for text adventures coded in their Zork Implementation Language, which allowed for easy porting of their games to all kinds of machines. While the games written in said Zork Language have been available since 2019 (here's Zork I and here's The Hitchhiker's Guide to the Galaxy), the interpreters for these files were not. Now, "Infocom's original Z-code interpreters for various home computers of the 1980s" is up on Github.
Jonathan Haidt is increasingly worried about boys, too, and previews a chapter of his book about The Anxious Generation. For girls the digital revolution which lead to a mental health crisis was all about social media, for boys, it was retreat from the real world through gaming. Haidt also launched a collaborative Google Doc for solving "the collective action problem" of regulating social media: Solving the Social Dilemma: Many Paths to Social Media Reform, here's the GDoc.
Relatedly, the attorneys general suing Meta have unsealed complaints in their lawsuit and claim that Meta designed platforms to get children addicted: "Instagram and Facebook parent company Meta purposefully engineered its platforms to addict children and knowingly allowed underage users to hold accounts, according to a newly unsealed legal complaint. The complaint is a key part of a lawsuit filed against Meta by the attorneys general of 33 states in late October and was originally redacted. It alleges the social media company knew – but never disclosed – it had received millions of complaints about underage users on Instagram but only disabled a fraction of those accounts. The large number of underage users was an 'open secret' at the company, the suit alleges, citing internal company documents."
Andrew Przybylski in Oxford, once again, finds no evidence screen time is negative for children’s cognitive development and well-being "in a study of nearly 12,000 children in the United States".
Meanwhile, in "an analysis of 33 studies which use neuroimaging technology to measure the impact of digital technology on the brains of children under the age of 12" with "more than 30,000 participants" found that "screen time leads to changes in the pre-frontal cortex of the brain, which is the base of executive functions such as working memory and the ability to plan or to respond flexibly to situations. It also finds impacts on the parietal lobe, which helps us to process touch, pressure, heat, cold, and pain; the temporal lobe, which is important for memory, hearing and language; and the occipital lobe, which helps us to interpret visual information." While those researchers also identified positive outcomes, the majority was bad.
Another meta-analysis showed that "a clear link between disordered screen use and lower cognitive functioning". And recently, a study found that Screen Time at age 1 delays communication and problem-solving developmend by 2 to 4 years. These are only the most recent studies i just fished out of my link collection. I already wrote about why i doubt Przybylskis work here, and i'm doubtful about his new study too.
Related: The Washington Post thinks Schools should ban smartphones and Parents should help. I think that too, but i go way further than just that: In germany, 40 education researchers called for a moratorium on digital tech entering the classroom because "Tablets and Laptops make kids stupid". I signed their letter because study after study shows that screens badly influence the neurological development of kids, that reading books is far superior to screens and that handwriting is far superior to typing. All of this comes on top of research showing the mental health issues your young people. Additionaly, those researchers are only talking about kids age 12 and lower, and seriously, any parent giving screens to kids below that age without supervision is acting extremely irresponsible.
Does the Mere Presence of a Smartphone Impact Cognitive Performance? "A Meta-Analysis of the 'Brain Drain Effect'" finds... not much: "Working memory capacity was negatively impacted by the mere presence of a smartphone, while results for the remaining cognitive functions produced null summary effects."
In A Nutshell on How the Internet Breaks Your Brain
Substack Has a Nazi Problem: "An informal search of the Substack website and of extremist Telegram channels that circulate Substack posts turns up scores of white-supremacist, neo-Confederate, and explicitly Nazi newsletters on Substack—many of them apparently started in the past year. These are, to be sure, a tiny fraction of the newsletters on a site that had more than 17,000 paid writers as of March (...) But to overlook white-nationalist newsletters on Substack as marginal or harmless would be a mistake."
I know that Substack has a hands-off attitude when it comes to moderation, but they also have TOS against hatespeech, and some of the stuff The Atlantic found clearly crosses that line.
Fellow Substacker Marisa Kabas is organizing an collective effort to make the Substack management adress that problem and you can participate by filling out this google form: "The plan is to draft a collective letter that we'll all post on our respective Substacks on an agreed upon day in order to create a groundswell of attention to the issue." I signed up and will participate.
The Guardian thinks that a big oil producer presiding over Cop28 is a risk, and here's why: President Of UN Climate Summit Lobbied For Oil And Gas Deals Behind The Scenes and Cop28 host UAE has world’s biggest climate-busting oil plans, data indicates and ‘Brand Dubai’: how the city-state is using Cop28 climate talks to build influence.
The OPEC is using the Cop28 as a promotional tool, while Saudi Arabia has a secretive program to keep the world burning oil. It's a farce and i listen to pretty much nothing coming out of Cop28.
The Scale of The Telegraph's Climate Change 'Propaganda': "The influential newspaper featured ten opinion writers with links to the UK’s main climate science denial group."
History fades as rising sea levels slowly destroy Thailand’s temple murals: "Saltwater damage could see precious historical Buddhist artworks dating back hundreds of years slowly fade entirely from view".
Unexpected consequences: Large scale events are developing too much heat: Taylor Swift's Brazil concert was hammered by extreme heat. How to protect crowds at the next sweltering gig.
Unexpected consequences 2: Exploring unmitigated risks of Microplastics and Nanoplastics in Sextoys: "Both micro- and nanoplastics particles were generated following the abrasion test, with the 50 percentile diameters (D50) ranging from the anal beads at 658.5 μm, dual vibrator at 887.83 μm, anal toy at 950 μm, and external vibrator at 1673.33 μm. The material matrix of each product was analyzed using ATR-FTIR, with results identifying the anal toy as polyethylene terephthalate (PET), the anal beads as polyvinyl chloride (PVC), the external vibrator as a silicone blend (polydimethylsiloxane [PDMS]), and the dual vibrator as a rubber mixture (polyisoprene). After extraction, phthalates known to be endocrine disruptors were present in all tested sex toys at levels exceeding hazard warnings."
Now you know. Insert with care.
Being nice predicts higher life satisfaction: "conversations with strangers and weak ties, as well as simply greeting and thanking weak ties, predicted greater life satisfaction", something Truman Burbank, Forrest Gump and yours truly could've told you without any studies. In my online persona, i often come of as a somewhat rude guy. You should meet me in person. Good morning, and in case I don't see ya, good afternoon, good evening, and good night!
Your Eyes Talk to Your Ears: "Ears make a subtle, imperceptible noise when the eyes move" and now researchers can "actually estimate the movement of the eyes, the position of the target that the eyes are going to look at, just from recordings made with a microphone in the ear canal".
Fracking Eyeballs: "How an alliance between psychologists and advertisers at the turn of the 20th century taught us how to measure (and monetize) human attention."
Kristin Andrews writing on the study of consciousness in crabs questions human-, language-, or even neuro-centric views on consciousness. She proposes vision as a simple cognitive marker for conscious experience, but i'd go much further: A will to live indicates that there is at least some form of experience of the world and that it is better for that individual organism than non-experience. This would even include microbes into the realm of conscious animals, and if you think about consciousness not as something that switches on at one point, but emerges gradually on a continuum, then "a will to live" to me seems the most basic common determinator. Consider it the anti-freudian anti-deathdrive of consciousness.
When predicting other people shapes the social mind, then there's a collective dimension to theory of mind itself. If you aren't a hermit, then your thoughts are always directly connected to the social world and vice versa: The social world shapes your thoughts and your theory of mind shapes the social world.
"I have to translate the colours": "Imagine if every time you heard a hard 'a' sound, you saw a luminous cyan colour, or every time you heard 'k', you saw red. Now imagine translating meaning from those colours, rather than the sounds themselves. This is the everyday experience of a woman identified in the literature as 'VA'."
Language heard while still in the womb found to impact brain development: "the research team found that the babies listening to the story in French showed an increase in long-range temporal correlations — all of a type that has previously been associated with speech perception and its processing".
Fede Alvarez' Alien: Romulus is set between Alien & Aliens. With nothing else known about the sequel this hints at a story about the colonization of LV-426 and its gory end.
Nosferatu from Robert Eggers (The Witch, The Northman) comes for christmas 2024. Can't wait.
Godzilla Minus One earns perfect Rotten Tomatoes score. I'm watching it tonight and i'm looking forward getting myself atomblasted by the original kaiju. I already loved Shin Godzilla and this looks even more promising.
Trailers worth watching: Night Swim Trailer 2, Club Zero, Society of the Snow, The Boy and the Heron - Final Trailer, The Animal Kingdom.
For Thanksgiving, Louie Zong made a satirical album in the style of Vince Guaraldis Peanuts-soundtracks: It's Black Friday, Charlie Brown, and it features tracks like Cyber Monday Blues, Buyer's remorse or Snoopy vs Capitalism.
Weird Al Yankovich in his Spotify year-end video: "It’s my understanding that I had over 80 million streams on Spotify this year. So, if I’m doing the math right that means I earned $12, enough to get myself a nice sandwich at a restaurant. Thanks for the sandwich."
The Mysterious Music on the Compilation Gespensterland Captures an Eerie Sound That Evades Definition: The german label Bureau B has put out the new compilation Gespensterland featuring synth tunes that pretty much sound like what you get when you mash the concepts of german angst with ambient, retro synth pop and krautrock.
And speaking of Bandcamp: Bandcamp Just Trying To 'Keep The Lights On' Following Epic Sale, Layoffs. I'm thinking about bulk downloading my 600 purchases there in the future.
In the last roundup, i posted about "the brightest flash of light ever seen in the night sky", now we have the most powerful cosmic ray since the Oh-My-God particle with an extremely high-energy particle detected falling to Earth. I think the universe is trying to say something.
Before we launch libraries on the moon, how about we don't close all the libraries here on earth? That would be something.
as always: I had a blast reading this.
Can't wait for Furiosa!