NON AI STUFF
Simon and Art Garfunkels Sound of Silence is a real thing: Silence Is a ‘Sound’ You Hear, Study Suggests: "In a paper published Monday in the Proceedings of the National Academy of Sciences, researchers used a series of sonic illusions to show that people perceive silences much as they hear sounds. While the study offers no insight into how our brains might be processing silence, the results suggest that people perceive silence as its own type of 'sound', not just as a gap between noises." (...) Rui Zhe Goh, a graduate student in cognitive science and philosophy at Johns Hopkins University and one of the scientists involved in the study, described a koan that he likes: “Silence is the experience of time passing.” He said he interprets that to mean that silence is “an auditory experience of pure time.”
RIP Kevin Mitnick, once the "worlds most wanted hacker". His book "The Art of Deception" provides some of the most valuable lessons in hacking, which is primary about social engineering and only secondary about the command line. Hackers are much more likely to target your trashcan than your computer, if they really mean it.
The guys from MSCHF are crowdsourcing a digitally handdrawn remix of "The Bee Movie" in which users visiting the site redraw one frame per user. The Bee-thing is already done, but they're doing John Wick 4 now, so scribble ahead. Reminds me of the Star Wars: Uncut-projects where users reshot scenes which then where strung together into a Star Wars-remix and to which yours truly contributed a Jawa that did not make it into the final version, and the fantastic Bartkira-project, in which illustrator Ryan Humphrey asked hundreds of colleagues to redraw every single page of Otomos famed manga with Simpsons characters instead of mutant kids from Neo-Tokyo. This also got turned into a short animated clip you can watch below, and when this was new, i exported all thousands of JPGs, turned them into readable CBR-files and made torrents for them which are all dead now.
Randall "XKCD" Munroe seems to get into typography lately. Alphabet Notes are some fun critical remarks about the typography of the alphabet ABC-sequence, and in Geohydrotypography, i just learned that the Atlantic Ocean expands by 10pts per month. One point is the standard unit for font sizes and line thickness.
While AI-doomers loose their sleep over rogue AI killing everyone in an instant, I'm way more worried about the existential threat posed by the environmental triumvirat of death: Plastic pollution, climate change and animal extinction. Here's Elizabeth Kolbert in The New Yorker telling How Plastics Are Poisoning Us. It's fine.
Surreal Photos of People Chilling in Bomb Shelters to Beat the Heat In China
I blogged about this phenomenon some years ago and now it's confirmed: The Ocean Has Changed Color Over 20 Years, Study Determines: "the oceans are becoming greener over time due to climate change (which) is shifting the color palette of the seas by disrupting tiny marine organisms called phytoplankton". The deep ocean blue we know today may be a thing of the past in a hundred years.
Jim Hansen at it again: ‘We are damned fools’: scientist who sounded climate alarm in 80s warns of worse to come. He also called the Paris agreement a "fraud", and I agreed back then and i still do now, because it's not legally binding.
The NYT has more on the whitest white pigment i already linked to here: To Ease Global Warming, the Whitest of Paints: "The paint’s properties are almost superheroic. It can make surfaces as much as eight degrees Fahrenheit cooler than ambient air temperatures at midday, and up to 19 degrees cooler at night, reducing temperatures inside buildings and decreasing air-conditioning needs by as much as 40 percent. It is cool to the touch, even under a blazing sun, Dr. Ruan said. Unlike air-conditioners, the paint doesn’t need any energy to work, and it doesn’t warm the outside air."
DNA for data storage is anything but new: In 2016 Microsoft stored OKGos musicvideo for This Too Shall Pass in genes, 2018 Massive Attack packed their Mezzanine-album into strains of DNA for the 20th anniversary of the album and one year later, scientists packed all of Wikipedia into genes with a DNA writer. Now researchers developed BacCam which "enables the direct-to-DNA storage and retrieval" of 1-bit images in 8x12 pixels which is not much, but it's writing the image information directly to DNA, just like a camera would.
Kathryn Finch in the excellent Blood Knife Magazine is celebrating 90 Years of Freaks. I'll never forget the living torso, played by Prince Randian, crawling through the rain carrying a knife with his teeth.
They found me: Archaeologists Discover Ancient 'Portal to the Underworld' In Cave Containing Skulls, Weapons
Alan Becker, the guy behind the famous Animator vs. Animation-shortfilm, now made the intellectually exhilarating Animation vs. Math and here's a complete over-analysis of all the mathematics i still don't really understand.
The Blob Toy -- That sand game, but with whibbly whobbly blobs.
Archive.org has thousands of LEGO Building Instructions, including the legendary classic 928 Galaxy Explorer. From the newer models, I'd expect myself to get lost in the Death Star or the Star Destroyer or the NASA Apollo Saturn V or the Space Shuttle or the Atari 2600 instructions, but no. My favorite is the Birdhouse.
Architecture for dogs including instructions and downloadable blueprints to build some sophisticated stuff for your furball.
Diaorama Cats -- giant felines meet miniature trains: "We operate a Japanese cat shelter while running a diorama cafeteria." You have not lived your life to it's fullest until your model train bumped into a giant cat butt.
The World’s Writing Systems "presents one reference glyph and basic information for each of the world’s writing systems. It is the first step of the Missing Scripts Project, a long-term initiative that aims to identify writing systems which are not yet encoded in the Unicode standard."
Legal Lullabies: "Lull yourself to sleep with the soothing white noise of your favorite tech giant’s terms of service. Close your eyes, drift away, or scroll down to join the 1% of technology users that claim to have read em’ from beginning to end."
You can vote until today for the Tiny Awards, a celebration of the small, playful, and heartfelt web.
AI STUFF
AI-race is heating up with Cupertino entering the ring: Apple Preps Ajax Generative AI, ‘Apple GPT’ to Rival OpenAI and Google. Apple chasing OAI/MS and Google gonna be interesting to watch.
AI-Johnny Cash sings Barbie Girl
Meta released their Open Source Llama 2 and if you read this blog for a while, you know that i'm not a fan of open sourcing AI-tech for all kinds of reasons, from self-radicalization to the vague may-or-may-not-materialize existential threats. I already laid down my argument in Limitations of alignment and AI-regulation, where i compare open sourcing AI-tech with open sourcing weapons, and here we are: WormGPT - The Generative AI Tool Cybercriminals Are Using to Launch BEC Attacks "was allegedly trained on a diverse array of data sources, particularly concentrating on malware-related data". While this one is "just" good at generating phishing emails, this blackhat stuff will just improve and the methods will get more and more sophisticated: PoisonGPT: How to poison LLM supply chainon Hugging Face: "We will show in this article how one can surgically modify an open-source model, GPT-J-6B, and upload it to Hugging Face to make it spread misinformation while being undetected by standard benchmarks." Given that AI might be the most powerful technology ever developed, we may be in for a ride. And this is just boring InfoSec stuff. I'm looking forward to the one-click models that mimic your special one to make you spin in circles.
VBox – AI-powered radio for musical exploration: "VBox is an AI-powered radio for musical exploration and group vibrations. Through the Large Language Model, VBox captures the various abstracted understandings of the song played in the form of texts and lets you travel down one of these rabbit holes to find more of what is hard to describe. (...) VBox derives its intelligence from OpenAI’s language models, which are prompted to analyze the playing track and extract its various abstract musical, cultural, and contextual properties. The same models are also used to find the tracks that share similar properties. The identified new tracks are streamed through Spotify and played through the VBox’s speakers."
Researchers looked into Personality Traits in Large Language Models. From the abstract: "We find that: 1) personality simulated in the outputs of some LLMs (under specific prompting configurations) is reliable and valid; 2) evidence of reliability and validity of LLM-simulated personality is stronger for larger and instruction fine-tuned models; and 3) personality in LLM outputs can be shaped along desired dimensions to mimic specific personality profiles." And from the paper: "This works demonstrates that it is possible to configure an LLM such that its output to a psychometric personality test is indistinguishable from a human respondent’s, and that it is possible to control the personality and in turn the output from such a model in a principled way. Furthermore, this work provides a complete pipeline to a) reliably and validly probe personality traits that may be perceived by humans in LLM output; b) identify the positive and negative emotions and other psychological factors they may be correlated with; and c) provide mechanisms to increase or decrease levels of specific LLM-synthesized traits." This is the stuff Daniel Dennett and Yuval Noah Harari warn about, and which i think will have the biggest impact of this technology: Automated simulations of synthetic personalities indistinguishable from people. The success of Character.AI is a pretty good indicator for this and the psychofrankensteinian phase of humans has begun. I have a bad feeling about this, especially regarding open source.
While the actors strike in Hollywood is about many things, it is the AI-proposal by the studios which gained the most attention. In short, the studios want to scan extras working as actors in the background for a one time fee and then use their digital likeness forever. Ethan Hunt himself lobbied for the unions and seems to worry about developments in mimetic AI. Relevant too is this story from April in which the CEO of Metaphysic -- the guys behind Deepfake Tom Cruise and that AI-Elvis on America's got Talent, about which I've blogged before -- registered his digital likeness for copyright. I don't know if the registration went through or how this is going, but i'd bet that we'll have a copyright for persons soon. In germany we do have a "Recht am eigenen Bild", which is not a copyright but a mechanism for citizens to maintain their privacy People compare this to the recent Black Mirror-episode "Joan is awful", but i'm much more reminded of the real life story of when Jet Li rejected a role in "The Matrix" because Warner wanted to bodyscan his martial arts moves and keep the rights. As a former union guy, the actors have my solidarity.
Somewhat related: In SHOW-1 and Showrunner Agents in Multi-Agent Simulations, researchers developed a TV-Show-generator building on the ChatGPT-powered environment for autonomous agents i blogged about here. For this work, they "collected a comprehensive dataset comprising approximately 1200 characters and 600 background images from the TV show South Park" and dreamboothed a Stable Diffusion-instance for character generation and developed a whole pipeline to generate complete episodes including dialogue and plot. The future gonna be a billion episodes of your favorite show generated by users and licensed to AI-companies, all competing for attention on whatever social network will be the viral hotspot then. Meanwhile, the rich will enjoy handcrafted analogue theatre written by excentric human artists and pay those new jester-artists millions of bucks for the real shit.
Deepfake videos prompt false memories of films in half of participants, but not at a higher rate than simple wrong text descriptions.
On the Origin of LLMs: An Evolutionary Tree and Graph for 15,821 Large Language Models. Explore this thing here.
‘A relationship with another human is overrated’ – inside the rise of AI girlfriends -- Just wait until someone hooks up customizable romantic AI with porn-synthesis. We ain't seen nothing yet.
An "AI build tool on minecraft" so you never have to play Minecraft again.
"Created by Aria Xiying Bao and Yubo Zhao at the MIT’s School of Architecture and Planning, Narratron is an interactive projector that augments hand shadow puppetry with AI-generated storytelling."
Kyle Goodrich built the DreamGenerator , a camera that "leverages Stable Diffusion and ControlNet". Filters on video apps are so last year.
Justin Alvey " jailbroke a Google Nest Mini so that you can run your own LLM’s, agents and voice models."
I'm not into Hiphop (except sometimes) but these AI-generated MF DOOM songs sound good to me and miles ahead from the cheap AI-Drake.
BillyBassGPT: "I put myself into a fish using GPT".
1980s horror b-movies relevant to our modern times, generated by ChatGPT, movieposter'd by Midjourney.
She’s Barbie Girl in a Nuclear World: "Death in plastic, its fantastic."
I really dig those vertical expansions of movies with Photoshop Generative Fill:
If silence is indeed a sound then why are deaf people called deaf?