Automatic Swatting as a service
Plus: Max Tegmark at Lex Fridman / AI-Clones of Weeknd and Drake / AI And The Dark Side Of Anthropomorphism / 100 gecs at Boiler Room / Florida Woman vs Damien Hirst and much more.
A Computer Generated Swatting Service Is Causing Havoc Across America:
Motherboard has found, this synthesized call and another against Hempstead High School were just one small part of a months-long, nationwide campaign of dozens, and potentially hundreds, of threats made by one swatter in particular who has weaponized computer generated voices. Known as “Torswats” on the messaging app Telegram, the swatter has been calling in bomb and mass shooting threats against highschools and other locations across the country. Torswat’s connection to these wide ranging swatting incidents has not been previously reported. The further automation of swatting techniques threatens to make an already dangerous harassment technique more prevalent.
Years ago when we had this abomination of internet drama called Gamergate, i speculated about automatic swatting, in which “trolls“ would simply scale swatting by distributing calls to the police with scripts. I didn't predict the AI-voices, but i think i nailed that one.
On a sidenote: I put “trolls“ in quotation marks because this is orwellian newspeak invented by techutopian netactivists for criminal psychopaths, and i think we should name things as they are and not find cute cozy new words to describe such behavior, just because they use new tools. A murderer with an AI is still a murderer, and a psychopath with a internet-connection is still a psychopath, “swatting“ is attempted murder and a guy who sells AI-voice-swatting is a hitman. (It takes away a lot of techno-romanticism when you call things by their real name, doesn’t it?)
However, now we have people selling “swatting“ with automatic AI-voices, tomorrow they’ll have an autonomous open source AI-system in place to automate this process and open the whole world of hacking to it. Maybe not today, but within a few years, maybe months, you’ll be able, as a skilled hacker, to hide indirect prompt injections on the Wikipedia-page of some person you don’t like and send the police to the homes of every visitor of that page, or everybody who submits Wikipedia-edits of a certain kind.
Simon Willison just published a piece called Prompt injection: What’s the worst that can happen? He not only talks about how people might be manipulated by hacked AI-assistants who follow hidden prompts, which is an attack vector i desribed in my piece on the AI risk of a synthetic theory of mind, but many other hacking methods via AI-systems piped into loops that users have no real oversight in autonomous AI-systems:
How certain can you be that ChatGPT running multiple plugins won’t produce a link that exfiltrates data, based on the user asking for a summary of their latest email messages through one plugin, which results in an attack like this one triggering actions in others?
I imagine there are far more sophisticated and malicious attacks that could be devised here by suitably motivated attackers.
As I wrote in my piece on the ChatGPT-Sims: “I don’t want to be near an autonomous AI-agent that (…) stumbles on a website with a hidden prompt injection and going haywire in the process.“
Because, soon, with all those open source things floating around and API calls automatized by ChatGPT, this process might involve your voice which calls real cops carrying deadly guns to my home.
In a recent piece on LLMs and Phishing, Security-researcher Bruce Scheier wrote “We don’t know how to live in a world with a billion, or 10 billion, scammers that never sleep.” I suspect that’s true for any bad actor out there, including skilled psychopaths and their digital methods of attempted murder by cop.
Max Tegmark: The Case for halting AI-development
I still haven’t read Tegmarks Life 3.0, but I like this take on capitalism and AI, which is very compatible with my view of seeing organizations and corporations as a form of semi-intelligent systems:
The issue with capitalism and the issue with runaway AI have kind of merged now, because [Scott Alexanders] moloch is exactly the capitalist moloch that we have built: an economy optimized for only one thing — profit. That worked great back when things were very inefficient, as long as the companies were small enough that they couldn't capture the regulators. But that's not true anymore.
They keep optimizing, and now these companies realize that they can make even more profit by building ever more powerful AI, even if it's reckless. This is moloch showing up and I just want to anyone who has any concerns about how late stage capitalism having gone a little too far — you should worry about super-intelligence, because it's the same villain in both cases.
I’m not the biggest anticapitalist in the world, but this rings true, and neoliberalism is just an effort to produce organizational AI without ethics or safeguards.
And while we are at Lex Fridman: Here’s two hours with the wonderful Simone Giertz: Queen of Shitty Robots.
Synthetic AI-Drakeweeknd sounds just as lame as the Original
So, there’s a viral track making the rounds, AI-generated with synthetic voices by Drake and The Weeknd. Drake seems pissed and Universal Music is urging streaming services to “help block the AI programs that are creating new songs based on the music of UMG artists”, which is funny because I’m kind of sure that training of AI-systems which can generate music does not involve sucking down the whole catalog on Spotify, but that’s beside the point.
The point also is not about how Drake and The Weeknd never made one worthwhile piece of music in their mediocre lives, or how this track surely doesn’t “go hard“ — this is just common wisdom and everybody with an ear or two knows that. You can listen to the track here, but be warned: it sucks — hard.
For a less sucky track to give you an example of what AI-music can do these days, here’s Michael Jackson singing a track by Bruno Mars, here’s Kanye West singing Oasis’ Wonderwall and, in good old absurdist music-mashup fashion, here’s West singing Careless Whisper. The quality of the last one is not quite there yet, but like in image synthesis, these flaws will go away. (For actually good music, visit my Side-Substack at, well, GOOD MUSIC.)
The point is that artistic styles and all other cultural properties are interpolatable now, and filesharing is an easy peazy thing to regulate compared to this. The stochastic and interpolative nature of generative AI goes against the principles of american and european copyrights which require individual, identifyable works by natural persons to take hold. We have no idea of how copyrights can be applied to an interpolatable latent space in which you have no clue how much Data from the music of Nirvana goes into a synthesis prompted by the word “Grunge“. As lawyers say: “It depends“ and this stuff is decided on a case by cases basis anyways.
Collecting society and performance rights organizations like GEMA or BMI have no strategy to make sense of stochastic mashups that work by atomizing existing culture and distributing it anew following weights and prompts. Even when Copyright-holders find a solution you will always have AI-warez full of weights for every song you can imagine, ready to mash them into bits’n’pieces, offering new ways of exploring and enjoying culture of the past, like imagining a Weeknd/Drake-track that never existed.
AI-Coversongs-as-a-service are a thing already and while i’m sure this website will not be online for very long, i’m also very sure that we’ll talk about copyright, interpolation and generative AI for a very, very, very long time to come.
update 04/18: For what it’s worth, the Weekndrake-mashup has been pulled from Streaming services: “The song, called Heart on My Sleeve, has been removed from TikTok, Spotify and YouTube for ‘infringing content created with generative AI’“.
Related:
Links
Luke Winkie at Slate writes about how LLMs only replace bullshit jobs in journalism: What time is the Super Bowl? A.I. can take that one, and all the other menial journalism jobs: “tools like Buzzy the Robot will inevitably eliminate a certain type of menial journalism job. But I’d argue that those positions probably shouldn’t have existed in the first place, because—plain and simple—they fail to meet the basic criteria of being either a job or journalism.“ — I had roughly that mindset when writing about how AI is Synthesizing the Businessman Smile.
Elon Musk Says His AI Project Will Seek to Understand the Nature of the Universe — Sure buddy. The guy who neither understands free speech nor social media besides shitposting wants to build Deep Thought. Looking forward to the Doge-memes.
Brooks Riley writing in the always excellent 3quarksdaily about The Great Pretender: AI And The Dark Side Of Anthropomorphism, writes about the points i was making in The AI-risk of a synthetic theory of mind:
”anthropomorphisation is arguably hard-wired into our minds and might have an evolutionary basis (Zlotowski et al. 2015). Even if the designers and engineers did not intend the robot to exhibit social signals, users might still perceive them. The human mind is wired to detect social signals and to interpret even the slightest behaviour as an indicator of some underlying motivation.In other words, one cannot help but react to certain aspects of a dialogue, because the chatbot has been outfitted with a toolbox of human reactions and phrases to help it navigate the transaction called conversation.
With the possible exception of the late gorilla Koko, we have never been able to seriously communicate with animals—or any other entity but ourselves. That we can now do that—and with a stunningly sophisticated simulation at that—is a profound turn of events which may be far more consequential than we now realize, distracted as we are by ChatGPT’s intellectual sleights of hand.”
PrankGPT - Prank people with AI-voices by phonecall. This is something you really want to see in the age of automatic swatting and Fake-AI-voice-kidnappings.
AI can’t read or write ASCII-Typography — Janelle Shane provides no explanation for the weird inability of LLMs to read pretty easy ASCII-typo. I suspect it’s something about how these letters often are used as delimiters in code, but maybe it’s a more mundane problem.
Auto-GPT Unmasked: The Hype and Hard Truths of Its Production Pitfalls — Indepth look at the pitfalls and costs of looping autonomous AI.
Teenage-AGI — more autonomous AI-experiments based on AutoGPT.
Francisco Toro writes about Our Deep Blue Moment on Persuasion: “The fear in that Deep Blue moment was that AI would displace human mastery. It didn’t. It turbocharged it. This, I think, is the basis for an optimistic read of our AI-enabled future. In some areas, humans may just be left in the dust by the machines. But in many others, what we’ll see is AI becoming deeply woven into people’s professional lives.“
Fantasy Internet Simulator — ChatGPT dreaming up webpages from circa 1995, mashed with any world you like: “ancient rome internet? what about the internet from star wars?“
A Method for automatically animating children’s drawings of the human figure — I’m old enough to remember The Monster Project in which professional illustrators rendered kids drawings into kickass colorful illustrations to everybody’s delight, and it’s a bit sad to see this beautiful thing being automatized.
GPT Unicorn — Daily SVG-Unicorns generated by GPT4. It’s going great.
Last week, Native Foreign, in collaboration with OpenAI, released the animated short 匚尺丨ㄒㄒ乇尺乙, a not-very-interesting, very flat looking and slow animation of Dall-E-generated visuals. The WaPo is using that opportunity to write about the state of AI in the movie industry and the pitfals of the tech: AI technology is coming to Hollywood. The filmmaking town isn't ready.
Replacing my best friends with an LLM trained on 500,000 group chat messages. You get an AI, you get an AI, everybody get’s an AI.
Machine Learning Street-Talk: Can AI Create Real Art?: “Dr. Tim Scarfe and Sam Roffi discuss the nature of art, its value, meaning, and its connection to subjectivity and objectivity. Their conversation explores generative art created by AI systems and whether it can have real meaning or value.“ — Spoiler: No, AI can’t create real art.
In the New Yorker, Cal Newport asks What Kind of Mind Does ChatGPT Have? Spoiler: None at all, and AI-Systems are not really intelligent. This is why i prefer to call them “stochastic libraries“. (I’ve written about that term here, in german.)
Can Intelligence Be Separated From the Body? — Intelligence is a property of a nervous system and comparing current LLMs to this is a fascinating philosophical exercise, but also very funny.
Photographer admits prize-winning image was AI-generated — I’m blown away regulary by the stuff that’s happening with post-photography and some creatives call image synthesis “photography in latent space“. Given that i don’t think this prankster thought this through when submitting his (pretty good) AI-generated image to the more experimental section of this reknown photo-award. But sure, have your fifteen minutes.
Beautiful Tape Cassette Inserts on Flickr
Kevin Kelly: Dreams are the Default for Intelligence: “I have a proto-theory: That our brains tend to produce dreams at all times, and that during waking hours, our brains tame the dream machine into perception and truthiness. At night, we let it run free to keep the brain areas occupied. The foundational mode of the intelligence is therefore dreaming.“
The wonderful soundtrack-blog El Diabolik posted it’s first episode in two years featuring weird, psychotronic filmmusic featuring songs from Hells Angels on Wheels, Bullit, Airport, Vixen and many others. This whole blog is a gold mine if you dig soundtracks to obscure midnight movies of yore.
Then this happened: Florida Woman Drives Rolls-Royce Into $3M Damien Hirst Work
The Battle Over Techno’s Origins — To understand this article properly, you have to know the decade old Frankfurt/Berlin-fight over “who owns Techno“. Berlin-Techno had its roots in Detroit, Frankfurt in Ibiza/Goa with Sven Väth on one side and EBM/Industrial with Talla 2XLC on the other. I’ve been there when this stuff happened and while enjoying the more melodic sound of Frankfurt (i’m original from that area), it was the fusion of that sound with Detroit Techno in Sven Väths legendary friday nights at the Omen that gave rise to what germans call “Schranz“ which dominated Techno for a long time and arguably still does today. The Hacker News discussion has some surprisingly good backgrounds on this story.
Daniel Keller reached out to Don Davis, who painted the fantastic 70s Space Colony images for NASA you’ve surely seen online somehwere, about commissioning new ones. “He’s down to it“ and I can’t wait.
Salman Rushdie Is Recovering, Reflecting, and Writing About the Attack on His Life
100 gecs at Boiler Room — Belongs to GOOD MUSIC, but everybody should know about 100 gecs because they are awesome.