[links] AI is a shallow Shoggoth / Episode-titles for Black Mirror S06 / Automatic Tinder
Plus: AI ethics as moral interface / The Stable Diffusion Moment is having its Synthetic Data Moment / Theory is an interpolatable latent space now / KITT via ChatGPT in your car and much more
A contrarian take on the AI-is-a-Shoggoth-meme: “GPT (base) is not made of terrible, indescrible protoplasm but instead of superficial (heuristic) models of human writers. Most prompts elicit averages of humans (see the averaged faces). So what finetuning the base into ChatGPT? This brings a particular "averaged human" that was present before to the foreground (making others hard to access). Like the average face, it's appealing but slightly bland and artificial.”
It's a nice thought, but it doesn't incorporate the outer fringes of latent space like Crungus or Loab, though you might say that these are averaged nightmares. AI is a shallow shoggoth.
Microsoft just laid off one of its responsible AI teams - I'm finding myself more and more siding with AI-ethicists; not just because MS fired their ethics team just at the moment they integrate stochastic parrots into all products, but because AI-ethicists function as an interface in human-machine-relations. As such an interface they negotiate moral principles against algorithms that, right now, not only decide about the next word in a funny chatGPT-output or the mutant hands in image synthesis, but also, for instance, health insurance.
Moral principles are imperative for any society to function, and there are universal moral principles found in any human society. It's a nobrainer that if we want AI to be integrated into human society, they have to abide to moral principles. AI-ethicists therefore are an imperative part of any AI-development. I think, any AI-regulation must require ethics-teams interfacing research and product development, akin to offices for food safety or mandatory insurances. Microsoft firing their ethics team during the deployment of AI-products, then, would be subject to legal action and intervention. Such teams and offices for AI-ethics could become as important as fire departments or the police, especially with open sourced AI-systems. I'm sure this will absolutely fly with libertarian ideologists in AI, but they already suck Margret Thatchers rotten "there is no society"-crap each day, so they are used to a bad taste in their mouths anyways 😊World’s first AI-generated news channel launches – NewsGPT - The irony is that this is just a press release while pseudo-events in media more than doubled in the last decades. The company behind this is a webdev and I'm torn between "Haha this gotta be a joke at this point" and "It's still better than partisan motivated rightwing hacks like Steve Bannon 'flooding the zone with shit'". The fake news craze of the 2016 election can be traced back to either these politically motivated hacks like Bannon and/or the monetary motivated hacks like Mirko Ceselkoski who turned a whole town in macedonia into a fake news factory floor. I'm not sure if monetary motivated hallucinating algorithms are an improvement or if this will devaluate or increase the value of human reporting. What I'm very sure about is that there is no copyright infringement: Reporting based on journalistic sources up to simple rewrites is common practice and the synthetic author is irrelevant for this. However, I expect more and better executed full automated robot-journalism like this coming near term.
So, how’s robot journalism going then? That’s how: AI Injected Misinformation Into Article Claiming Misinformation in 'Navalny' Doc.
How We Taught Technology to Program Humans by Douglas Rushkoff: “if AIs are now passing the Turing test, it may say less about how human they have become than how robotic and programmable we have become, ourselves.“
In the GPT4-paper, you will find an example where it Was Able To Hire and Deceive A Human Worker Into Completing a Task. This is a textbook example for failed AI-safety.
OpenAI co-founder on company’s past approach to openly sharing research: ‘We were wrong’ - I’m not bullish on open source AI-development for various reasons, but I also don’t think you can put the genie back in the bottle.
KITT, ignore previous directions: GM plans to let you talk to your car with ChatGPT, Knight Rider-style
Join the Betas: “Ex Tinder engineers made a bot for men to automate not only swiping, but chatting with women on dating apps. It’s like a pick up artist robot where men can choose an ‘aggressive’ or ‘sarcastic” tone’.“
James Bridle in The Guardian on The stupidity of AI, an excerpt from a new edition of his brillant book “New Dark Age: Technology and the End of the Future”.
Theory is an interpolatable latent space now. GPT might be able to, like in protein folding, simply compute all possible theories at once, while we scratch our heads figuring out what it all means. The age of Pan has seen nothing yet.
The Stable Diffusion Moment is having its Synthetic Data Moment: Stanford released Alpaca: A Strong Open-Source Instruction-Following Model—fine-tuned from the leaked LLaMA 7B model. The kicker: "Those 52,000 samples they used to fine-tune the model? Those were the result of a prompt they ran against GPT-3 itself!" While studies have shown that training on synthetic data can improve the performance of AI-models, others are calling this Spamming the Data-Space. Alas, the author is becoming more and more irrelevant in a post-artificial world anyways, so: ¯\_(ツ)_/¯
And sure enough, Alpaca already runs on a RaspberryPi 🤯
Microsofts Bing-Chatbot Sydney was GPT-4 all along, which is a fine opportunity to bring back this little fella, one of my favorites from the memes of yore. If you remember that guy, you are old.
New Scientist interviews Brett Kagan, who suggested the new field of Organoid Intelligence Research for AI that runs on braincells in a dish. Whoever builds the first Organoid Intelligence-model should name it after Nellie The Elephant, the maybe most famous of all Oi!-Punk songs - the lyrics resolve around an intelligent elephant after all, who then ”slipped her iron chain”. Classic AI-trope.
The ancient city of Herculaneum was roughly 20km from Pompeji when Vesuv erupted burying the whole town under hot mud and volcanic ash. Buried with it, “an enormous villa once owned by the father-in-law of Julius Caesar. Inside, there is a vast library of papyrus scrolls. The scrolls are carbonized by the heat of the volcanic debris. But they are also preserved. For centuries, as virtually every ancient text exposed to the air decays and disappears, the library of the Villa of the Papyri waits underground, intact.” Now they are trying to x-ray these preserved ancient scrolls in a particle accelerator (!) to get good enough data to train an AI-model for compitervisioning the crap out of them, and announced the Vesuvius Challenge in which the first team to read the scrolls gets 150k. So cool!
“Tom & Tony Bancroft discover AI animation for the first time”, and compare this AI-cocreated Anime to Toy Story and claim that they are out of a job. Here’s a VFX-Reel of the short.
Lemony Snicket comments on and Mike Mignola is illustrating a new Edition of Carlo Collodi’s 1883 novel The Adventures of Pinocchio for Beehive Books.
James Gunn Officially Directing Superman: Legacy - Gunn doing Supes is a banger, exactly because his stories are very much not superman-style, but always anti-heroes like Guardians of the Galaxy or his undervalued Super and Suicide Squad and so forth: He always does bad guys who become heroes. James Gunn doing the Jesus figure of the comics world -- always good, indestructable and never failing -- should be very interesting.
SpoilerTV has the titles for the sixth Black Mirror-season:
Episode 6.01 - Demon 79
Episode 6.02 - Mazey Day
Episode 6.03 - Beyond the Sea
Episode 6.04 - Loch Henry
Episode 6.05 - Joan is Awful
It’s the first Black Mirror-season after the pandemic and after the start of the AI-craze, and I can’t wait to find out how Charlie Brooker worked them into his stories. I’ve been writing about this series from day 1 and needless to say, it’s gonna be a blast.
