Holographic Einstein is the new Literacy
GOOD LINKS 2024-01-27: ChatGPT is Muffit III / Massively Multiplayer Art / Nazis and Uncensored AI / Nightshade / Facial Recognition on DNA-Faces / Climate Fiction for Future Ancestors and much more.
Art magazine Outland has a new Gaming-issue out looking at the intersection of art and play, featuring stories about "agency as an artistic medium", Life in the Ludic Century or Massively Multiplayer Online Art. Interesting stuff throughout.
I remember blogging about a website which provided PDFs of famous spaceships as papercraft cutouts and because everybody and their mothers are doing papercraft spaceships nowadays, here's Papercraft Fish instead. It's the same but fish. And while i know that octopusses are not fish, i don't care enough to not show a papercraft octopus cutout template because, well, it has tentacles. Way better than spaceships.
"Modern Illustration is a project by illustrator Zara Picken, featuring print artefacts from her extensive personal collection. Her aim to is preserve and document outstanding examples of mid-20th century commercial art".
DJ Food has some images from the Turn On zoetrope exhibition in Luxembourg featuring "phenakistoscopes and zoetropes revolving around music and audio".
IIOANA – Aqua Park, "a live code-able video synth and coding environment that runs directly in the browser".
Neo-Nazis + 'Uncensored' AI: The founder of a Neo-Nazi group "designated as a terrorist organization in Canada" instructed his followers to use "uncensored Large Language Models" like dolphin-2.5-mixtral-8x7b-gguf, a fork of the open source model Mixtral 8 made by a guy going with the pseudonym of The Bloke. 404 Media goes on to seemingly refute the common argument that "but but but you can find all information to build a bomb on google and in chemistry textbooks too", but they don't get to the bottom of it: The information on how to build a bomb in a chemistry textbook is scattered and disconnected, where in an "uncensored" LLM it sits there readily joined together in the interpolated latent space, waiting to be found by a prompt. This is why i'm against open sourcing automatized knowledge interpolators. However, "Mistral, The Bloke, and Replicate did not respond to a request for comment", because ofcourse.
Kyle Orland is not convinced that an AI wrote that hour-long “George Carlin” special, but George Carlin Estate Sues Podcasters Over AI “Comedy Special" anyways. I'm not convinced by his arguments, and want to remind him of the possibility of human/AI colab in which an LLM finetuned on Carlins routines puts stuff out and a human selects and edits for kicks, feeding that into a voice cloner et voilá. Sure they said they "trained" a model, but ofcourse that's just finetuning. Also, I don't see the point of faking AI-production for some joke with a punch line or an art project or whatever, when that fake is going on for two whole years. I'm very sure that if that Dudesy-crap would fake it's AI, they would've already come out with some Gotcha at the time OpenAI revealed ChatGPT, because that would've created maximum impact for their thing. I think "sloppy fun and thoughtless AI production" is a much more simple and likely explanation here, but maybe i'm wrong.
(update 28.1.24: I’m wrong: “Danielle Del, a spokeswoman for Sasso, said Dudesy is not actually an A.I. ‘It’s a fictional podcast character created by two human beings, Will Sasso and Chad Kultgen,’ Del wrote in an email. ‘The YouTube video ‘I’m Glad I’m Dead’ was completely written by Chad Kultgen.’“)
ChatGPT is an engine of cultural transmission: "'Gopnikism' (...) implies that these models are incapable of hallucinating falsehooods (...) An LLM isn't doing anything differently when it ‘hallucinates’ as opposed to when it gets things right", because it has no connection to base reality. I wrote about Alison Gopniks model of thinking about LLMs in Wishful Mnemonics - On Stochastic Libraries, and her framework is still the most helpful when it comes to LLMs. Also from Henry Farrell, a piece on humans tendency to see personality in LLMs: Kevin Roose's Shoggoth. Good stuff.
New Theory Suggests Chatbots Can Understand Text: By scaling foundation models, chatbots develop ever increasing "text-understanding skills" and can combine them, leading to generalization and the (disputed) phenomenon of emergence, the ability to solve tasks not present in the training data. The researchers say, this puts the "stochastic parrot" to rest. The math is flying over my head here, but i don't find it surprising that increasing training data increases skillsets, but i also find the term "understanding" misapplied here. A chatbot might "understand" text, but that understanding is still disconnected from reality, which is what understaning ultimately means in context of intelligence: Having a working model of stuff in our head (smell, haptics, look, colors, form, sound, emotions, memories, location etc etc), not just text and tokens relating to that stuff.
Are Large Language Models Conscious? Interesting article that makes a crucial mistake by conflating the absence of social behavior and language capacity in feral children with absence of consciousness. Feral Children are rare cases of kids being raised in the wild by wolves or other animals, and who lack human traits like sociality. The article wrongly declares them "linguistic isolates", but they are first and foremost social isolates, and language absolutely takes a second row here, as many cases of feral children show that were able to learn language and get, at least somewhat, integrated into human societies. Also, i see no reason to believe that language is a necessary prior for consciousness, depending on how you define consciousness. If you, like me, see consciousness as an agentic composer of brain activity which guides our attention towards stuff, then language is not necessary for that, and every predominantly visual-conceptual thinker (like me) can confirm that. Or as i like to say: Language is downstream from consciousness. Which also means that consciousness is at maximum only loosely connected to language use, and is something that can't emerge from language "understanding" alone, which might or might not be present in Large Language Models.
Interesting finding regarding human-AI colaboration: Superhuman artificial intelligence can improve human decision-making by increasing novelty. They analyzed "5.8 million move decisions made by professional Go players over the past 71 years (1950 to 2021)" and found that "humans began to make significantly better decisions following the advent of superhuman AI", namely the advent of AlphaGo, which beat Go world-chamion Lee Sedol back in 2016.
Forbes on a new wave of kids toys powered by ChatGPT. I know next to nothing about the psychology of developing children, but i'd suggest their naturally occuring imaginary friends they project into stuffed animals are something fundamentally different than a toy that comes with an "artificial personality" built in, but maybe i'm also too critical here because after all, Muffit is back.
AI Snake Oil dissects the copyright dilemma of the GenAI-industry and concludes that Generative AI’s end-run around copyright won’t be resolved by the courts. I disagree with that headline but i do agree with their reasoning, which goes like: Because output similarity can be easily fixed, the focus on plagiarising output is wrong headed, as the real harm is competition of AI systems in the same markets with artists and writers whose works are also used in the training data. This is precisely right, and this is why i think that compensation for training data should be mandatory, even when it means that Marc Andreessen has an emotional breakdown from paying people contributing their work to his product.
Nightshade v1.0 has been released, a tool that "transforms images into 'poison' samples, so that models training on them without consent will see their models learn unpredictable behaviors that deviate from expected norms, e.g. a prompt that asks for an image of a cow flying in space might instead get an image of a handbag floating in space." Interestingly, Alex J. Champandard points out that "Nightshade and its sibling Glaze are technological measures as defined by the DMCA that control access to a work (to ensure proper licensing), and circumventing them would be a violation of 17 U.S. Code §1201", which means that adversarial attacks can be applied as a new form of DRM.
Hologram lecturers thrill students at trailblazing UK university: "AI (is) the new form of literacy (…) Proto has the technology to project an image of Stephen Hawking, or anybody, and make it look like he’s really there. We can hook it up to books, lectures, social media – anything he was attached to, any question, any interaction with him. An AI Stephen Hawking would look like him, sound like him and interact like it was him." I don't think businesspeople selling holograms of Stephen Hawking to universities know what the word "literacy" means. Also, because holograms of Stephen Hawking are the new literacy, Dall-E 3 didn’t let me generate Holo-Hawkings so i had to go with Holo-Einstein.
EU set to allow draconian use of facial recognition tech: "Last-minute tweaks to the European Union’s Artificial Intelligence Act will allow law enforcement to use facial recognition technology on recorded video footage without a judge’s approval".
Meanwhile, in draconian use of facial recognition tech: Cops Used DNA to Predict a Suspect’s Face — and Tried to Run Facial Recognition on It. Reminds me of the art of Heather Dewey-Hagborg, who collected DNA samples on the streets of New York and reconstructed their faces with software and then 3D-printed them, except she didn't use facial recognition on them to solve real crimes, because that would be a stupid thing to do.
I'm not sure if that will prevail, but at least for now, OpenAI's GPT store is already being flooded with AI girlfriend bots. But don't worry, even when they crack down on these, there will be enough shady websites to find a synthetic slave for your fictosexual practice.
Robocall with artificial Joe Biden voice tells Democrats not to vote. Open Source Voicecloning AI is going great!
Most Top News Sites Block AI Bots. Right-Wing Media Welcomes Them. While the headline is slightly misleading, and WIRED buries that fact at the end of the article -- in short: Many if not most rightwing outlet simply didn't prioritise blocking AI bots and did so when contacted by WIRED, which means they don't "welcome" them --, the article hints at an unintended consequence of the legal AI-copyright battles, in which partisan media outlets fight over influence in on datasets by robots.txt. However, the real place of this fight is licensing: Springer was the first large media house to license their journalism to OpenAI including for the use as training data, meaning that conservative leaning perspectives from tabloids like the german BILD-Zeitung, which is famous for spurring outrage and ignoring privacy rights of citizens, will have a prominent size in the training data of the worlds most used AI-system. I wrote about some of the possible outcomes of this in Populist Media Manipulation in times of Synthetic Journalism.
Iceland Has Its Own AI George Carlin Moment, Considers Law Against Deepfaking the Dead: "The nation of Iceland is considering restrictions on using AI to reanimate dead people after the national broadcaster there used AI to resurrect a beloved comedian for a music video". Regulating the resurrection of the dead sounds reasonable to me, with or without AI.
I just wrote a piece on the dawn of the AI-military complex, and here's the next step: "Mobile datacenters (in mesh networks) that can be taken by a helicopter and dropped (...) on a battlefield near all of those devices, the drones, then it can help process all the data from those sources. You can do AI at the site and figure out what’s the data that’s actually relevant for a given thing and then just send that data back" without relying on the cloud.
Lumiere is a new AI-video model from Google with some impressive results, especially it shows somewhat consistency in parts that are actually moving, and footsteps, for instance, don't look too much like whibbly whobbly worm stuff anymore, but pure GenVid still has a long way to go.
The Psychopolitics Of Trauma: Scott Alexander dissects our addiction to political outrage, often subsummized under the umbrella-term "doomscrolling", through the lense of Trauma and PTSD, and it makes sense: "Suppose that outrage addiction is, in fact, trauma addiction. That means the media ecosystem is a giant machine trying to traumatize as many people as possible in order to create repeat customers, ie trauma addicts. Combine that with the explicit, confessed desire on both sides to 'trigger' the other as much as possible, and you have a lot of very clever people all trying to maximize one another’s trauma levels. On the external level, that looks like weaving as strong a narrative of threat and persecution as possible and trying to hit people in their psychological weak points. On the internal level, it means making sure they replace their normal ability to update with a series of triggers that make them replace reality with pre-packaged stories about how the other side is innately evil and everything they do is for specific threatening and evil reasons."
Are Internet Subcultures the New Punk Rock? I don't know about you but while internet subcultures may or may not be the new punk rock, it also strikes me as odd to have "13 year old MAGAcommunists". 13 year olds should neither be thinking about communism nor trumpism, they should think about Susi and maybe puke from their first beers and get into, yes, punk rock. This is what teens at that age should do, dumping a big one on the rules set by their parents and explore their boundaries, at least sometimes, not getting into fucking politics. I guess this is what you get from a generation superdown on safety: They don't drive or fuck or drink or, you know, have fun, they get into politics, of all things, because playing armchair politics with edgy memes on socmed is a safe activity, compared to drinking, fucking and driving. I like Citarella and i'm following his work since he started out, but what he's documenting here is not punk rock, but the gamification of tribalism on socmed and a politication of adolescence that strikes me as highly unhealthy for any society.
Meta has not done enough to safeguard children, whistleblower says: "Russell, a 14-year-old girl from Harrow, north-west London (...) died from an act of self-harm while suffering from depression and the negative effects of online content (...) after viewing harmful content related to suicide, self-harm, depression and anxiety on Instagram and Pinterest (...) Zuckerberg had the tools at his disposal to make Instagram, in particular, safer for teenagers but the company has chosen not to make those changes. (...) It would take three months for Meta to carry out an efficient crackdown on self-harm content, Béjar added. 'They have all the machinery necessary to do that. What it requires is the will and the policy decision to say, for teenagers, we’re going to create a truly safe environment that we’re going to measure and report on publicly.'"
The Substack Verschlimmbesserung: Good take on the whole Substack-situation: "We have accepted that hate-oriented communities have a corrosive effect on our broader culture without accepting that big conversations have a calming effective on those radical ideologies and their followers. But we know that keeping people in diverse communities is good for them, and shunting them into homophily is unnatural and unwise."
Oh look Technorati is back: Feedle: Search and Discover Quality RSS Feeds from Thousands of Blogs and Podcasts. (Yes, if you remember Technorati you're old.)
"Grist’s Imagine 2200: Climate Fiction for Future Ancestors short story contest celebrates stories that offer vivid, hope-filled, diverse visions of climate progress. From 1,000 submissions, our reviewers and judges selected the three winners and nine finalists you will discover in this collection." Beautifully illustrated by Taj Francis, Carolina Rodriguez Fuenmayor, Stefan Grosse Halbuer, Mikyung Lee and Molly Mendoza.
Stealing Higher Ground: A Portrait of Rising Climate Gentrification: Orion talks to Katja Esson about her new film Razing Liberty Square about gentrification in Miami in times of rising sealevels with a focus on "Liberty Square, one of the first public housing developments for Black people in the country". Those public housing projects were built in the 1930s on high ground, and with rising sea levels, this ground is increasing in value, resulting in the destruction of public housing in favor of the development of expensive flats for the rich escaping the waters.
Greenwashing seems to be the main activity of the Fossil fuel industry these days: The propane industry is trying to dupe you: "The Propane Education & Research Council (PERC), a U.S. lobbying group, has spent nearly $30 million over the last two years on advertisements for the fossil fuel, according to data compiled by Drilled, a multimedia reporting project focused on climate accountability. The ads often promote propane, the vast majority of which is a byproduct of natural gas or crude oil refining, as a form of clean and renewable energy."
Nuclear power output expected to break global records in 2025. Some people won't like this but i prioritize climate change much higher than risk from accidents involving nuclear reactors, so I'm fine with that. Renewables would be preferable, but as long as they are not fully there, any measure in leaving fossil fuels in the ground is welcome, because time for carbon neutrality is running out and 2024 is an inflection point for the climate crisis.
More good news: EU fossil fuel CO2 emissions hit 60-year low: "'EU CO2 emissions have finally fallen back to levels apparent in my parents’ generation in the 1960s', said Isaac Levi, an analyst at Crea. 'Yet, over this time period, the economy has tripled – showing that climate change can be combated without foregoing economic growth.'"
New Religion is a 2022 J-Horror film by Keishi Kondo in which a call girl gets visits from the spirit of her lost daughter by being photographed by a customer, but with every photo taken, society around her falters and collapses. I didn't know this film until now that i read about the follow-up shortfilm Neu Mirrors and all of this sounds highly intriguing.
New Trailer for the surreal anarchic slapstick action comedy thing Hundrets of Beavers, in which "a drunken applejack salesman is thrust into the frigid wilderness and has to trap his way out to survive". Looks weird af but we'll see if it can hold up.
More Trailers worth watching: Disappear Completely (Mexican crime mystery thriller about a tabloid crime photographer who becomes so numbed from his true crime job, he's literally loosing all senses), Constellation (psychological scifi-drama on Apple+), Ripley (Tom Ripleys further adventures in stylish murder on Netflix), Hit Man (Richard Linklaters movie about a hitman saving a woman from her abusive husband), Shōgun (new good looking series adaption of James Clavells novel), Double Blind (Scifi-horror about a failed drug experiment after which you die when you fall asleep), Monkey Man (Dev Patel revenge movie set in India), You'll Never Find Me (Claustrophobic psychothriller about "an isolated man living at the back of a desolate caravan park visited by a desperate young woman seeking shelter from a violent storm"), Tuesday ("Death, in the form of a size altering macaw, sends Zora and Tuesday on an emotional journey about life, love, and death"), Immaculate (fairly good looking nun horror movie), Road House (Remake of the 80s Patrick Swayze action romp with Jake Gyllenhaal as an Ex-UFC-fighter breaking bones in Florida, looks like brainless but good fun), The Ones Who Live (The Rick & Michonne-Walking Dead spinoff and i know i know, The Walking Dead is a stupid soap opera featuring Zombies, but i have history with it: I was an early follower of the pretty good comics and watched the first 5 or 6 seasons or so with a continuingly sinking heart because they largely fucked it up, and this, honestly, looks a bit like return to form, reminding me much of the first season by Frank Darabont set in zombie infected Atlanta, so i'll give this a shot).
Godzilla Minus One Minus Color gets its own black and white limited edition action figure. Now i want black and white Mad Max Fury Road action figures. I would buy any Coma-Doof Warrior with a black and white flame-throwing electric guitar on a black and white Doof Wagon in no time. But Black and White Godzilla is neat too.
‘Elfquest’ Animated Series Based On Comics In Works At Fox: Besides Heavy Metal, Elfquest was the quintessential Fantasy-comic of the 80s. I only read a handful of issues but i'm happy to see them finally get the recognition they deserve.
RIP Norman Jewison, director of Jesus Christ Superstar, In the Heat of the Night, The Thomas Crown Affair and the scifi-classic Rollerball.
The Oscars nominations 2024 are out and here's my predictions. My prediction about my predictions: They're all wrong.
Best picture: Oppenheimer
Best actor in a leading role: Cillian Murphy – Oppenheimer
Best actor in a supporting role: Mark Ruffalo – Poor Things
Best actress in a leading role: Lily Gladstone – Killers of the Flower Moon
Best actress in a supporting role: Da’Vine Joy Randolph – The Holdovers
Best directing: Poor Things – Yorgos Lanthimos
Best animated feature film: The Boy and the Heron
Best adapted screenplay: American Fiction
Best original screenplay: The Holdovers
Best cinematography: Oppenheimer
Best costume design: Napoleon
Best documentary feature film: 20 Days in Mariupol
Best documentary short film: The Last Repair Shop
Best film editing: Oppenheimer
Best international feature film: The Zone of Interest
Best makeup and hairstyling: Poor Things
Best original score: American Fiction
Best original song: I’m Just Ken – Barbie
Best production design: Poor Things
Best animated short film: WAR IS OVER! Inspired by the Music of John & Yoko
Best live action short film: The Wonderful Story of Henry Sugar
Best sound: Maestro
Best visual effects: The Creator
NASA’s Scout System Just Predicted an Asteroid Impact. Here’s What That Means for Planetary Defense. It means that we can now also detect small-ish Asteroids some time before impact to make possible at least some evacuation. That thing here went down over Berlin where i live, here's a video.
A good skeet: More Mr. Nice Guy "hired Werner Herzog to fake the moon landing and he just did it for real like Fitzcarraldo and now Klaus Kinski is running around up there" which creates very images in my head.
All The Clocks, from seconds to millidays to centuries to Teenage Kicks by The Undertones. I love that you can visualize those clocks not just in various abstract patterns, but also Chet Baker.
All the Types of Science Fiction. All of them.
The man who collects lost pet posters: "Don Bolles is a veteran of punk with a one-of-a-kind collection."
Snowcraft - Building a LEGO Snowfort from giant LEGO snow-bricks. I don't think i mentioned this in my post about Superb Halos where i also told some anecdotes about growing up camping and skiing in the mountains, but i also regularly build snowforts and Igloos pretty much every winter, albeit not from snow Lego-bricks, while my parents constructed outdoor bars made from snow to have some schnaps in the cold.
Ode to Internet - V.90 56k dialup modem handshake for orchestra:
How different languages laugh online: "Laughter is universal, but lol is not". If you want to lol in german, use "lel", but that's associated with troll lingo and likely out of date because i'm old now. A simple "haha" or "hehe" or "hihi" or "hoho" or "höhö" or "hähä" will do too , where "haha" is standard laughter, "hehe" is a more sharp and cheeky one, "hihi" is more of a restrained and shy laughter, "hoho" is good manered laughing or Santa Claus, "höhö" is something like positively cynical, and "hähä" is mean laughter. Nobody says "hühü" though, likely because "huhu" already is used as an abbreviation for "hello" 😊
The flowers are blooming in antarctica "and it is good and I feel so happy! 🌷🌸🌹🌺 I am carbonfootprintmaxxing so much right now. I love burning gasoline! ⛽⛽🛣️🛣️ default text."
Tiktok failed to load.Enable 3rd party cookies or use another browser
This Tomb Contains 10 Mummified Crocodiles From 2,500 Years Ago. Crocodily Mummies, what's not to love? Here's the paper: Newly discovered crocodile mummies of variable quality from an undisturbed tomb at Qubbat al-Hawā.
Hunter-gatherers were mostly gatherers: "Early human hunter-gatherers ate mostly plants and vegetables, according to archeological findings that undermine the commonly held view that our ancestors lived on a high protein, meat-heavy diet."
Elle Cordova is the best:
Tiktok failed to load.Enable 3rd party cookies or use another browser
On Friday, the creators of the George Carlin voice-cloned AI special admitted through a rep to the New York Times that it was not written using a large language model. "'Danielle Del, a spokeswoman for Sasso, said Dudesy is not actually an A.I. It’s a fictional podcast character created by two human beings, Will Sasso and Chad Kultgen,' Del wrote in an email. 'The YouTube video ‘I’m Glad I’m Dead’ was completely written by Chad Kultgen.'” https://www.nytimes.com/2024/01/26/arts/carlin-lawsuit-ai-podcast-copyright.html