[links] Noam Chomsky on ChatGPT / 4chan leaks LLaMA / Roope Rainistos Post-Photography
Plus: Historical AI-analogies, CBS 60 Minutes on ChatGPT, intergenerational GPT-Jailbreaking, trailers for Silo and TMNT: Mutant Mayhem, and much more.
Roope Rainistos Post-Photography
I’m still blown away by Roope Rainistos AI-art. I wrote about his work a few months ago, and for a few weeks now he’s doing a post-photography project called Life in West America:
Drawing inspiration from the early days of American color photography, the collection combines the visual language of traditional photography with the limitless artifice of AI. We go on a road trip that is both nostalgic and futuristic, inviting the viewers to question their perceptions of reality. (…) Drawing inspiration from the early days of American color photography, the collection combines the visual language of traditional photography with the limitless artifice of AI. We go on a road trip that is both nostalgic and futuristic, inviting the viewers to question their perceptions of reality.
This stuff is abailable as NFTs too, if you’re into this sort of thing.
He’s also curating a group show now, and check out his Youtube, where he posted some pretty cool experiments in AI-animations, running the Beastie Boys and Smashing Pumpkins through the diffusion-grinder, with “An Illogical Dance“ being one of the finest examples of the artform yet.
A few days ago I wrote about how AI-systems are like lovecraftian Shoggoths, so here’s another historical analogy from Robert Seymour, The March of Intellect, ca. 1828, “where a jolly automaton stomps across society. Its head is a literal stack of knowledge — tomes of history, philosophy, and mechanic manuals power two gas-lantern eyes. (…) Like Frankenstein’s creature, birthed from a pick ‘n’ mix of exhumed organs and ossified science, the monster in this satirical cartoon is patchwork knowledge itself, practically applied and made widely available for the very first time.“
Noam Chomsky writes in the NYT about The False Promise of ChatGPT (non-paywall)
I agree with a lot, but not everything here. He says, for instance, AI-systems “are incapable of distinguishing the possible from the impossible (…) They trade merely in probabilities that change over time.”
Which, fine, but isn’t the human ability to know right from wrong not too just statistics, giving more weight to that which is possible? An adult still knows that the myth Santa exists, even though he's not real. He just updated his knowledge aquired as a child, which is, you know: “Trading merely in probabilities that change over time.”Yann Le Cunn writes in NOEMA-Mag about how AI Chatbots Don’t Care About Your Social Norms: ”The problem with chatbots isn’t that they are black boxes or that the technology is unfamiliar. (…) The problem is that they don’t care. They don’t have any intrinsic goals they want to accomplish through conversation and aren’t motivated by what others think or how they are reacting. They don’t feel bad about lying and they gain nothing by being honest. They are shameless in a way even the worst people aren’t — even Donald Trump cares enough about his reputation to at least claim he’s truthful. This makes their conversations pointless.”
I’d like to add to this that this lack of intrinsic goals is exactly the reason why AI can’t produce art by itself.Facebook's Powerful Large Language Model Leaks Online: ”The leaked language model was posted to 4chan.”
I went a bit viral with this on Twitter last week when the model leaked. Here’s a comment on Hackernews with Screenshots from 4chan: ”Two people in the same thread had access to the weights and verified that their hashes match to make sure that the model isn't watermarked. However, the leaker made a mistake of adding the original download script which had his unique download URL to the torrent, so Meta can easily find them if they want to.”
Here’s a high speed download. People got the smallest LLaMA model with 7B parameters to run on a single GPU and the output has, at least in one of the outputs, a kind of Bing vibe to it. Lulz, I guess.CBS 60 Minutes devoted their latest installment to Large Language Models and AI, two segments are on Youtube: ChatGPT: Artificial Intelligence, chatbots and a world of unknowns and overtime: ChatGPT and large language model bias
As A.I. Booms, Lawmakers Struggle to Understand the Technology: ”Tech innovations are again racing ahead of Washington’s ability to regulate them”. Related: ChatGPT broke the EU plan to regulate AI.
How to convince a large AI, according to smaller AIs, intergenerational GPT-Jailbreaking by Janelle Shane
AI Chatbot ChatGPT Mirrors Its Users to Appear Intelligent - In a MIT-paper, researchers describe LLMs as a “reverse Turing Test“: “What appears to be intelligence in LLMs may in fact be a mirror that reflects the intelligence of the interviewer, a remarkable twist that could be considered a reverse Turing test.“
Related: Introducing the AI Mirror Test. I had doubts that LLMs are a true mirror test.Self-Learning Agent for Performing APIs: “When given a task, SLAPA knows to search for the API documentation and learn all the information. Then he create API calls. If they don't work, he learns from his mistake and tries again. SLAPA's knowledge accumulates. He only needs to read documentation if you ask for a new API he's never encountered before; otherwise, he already knows it.“
What can possibly go wrong, as they say.Scammers are now using AI to sound like family members. It’s working.
Related: Elevenlabs introduce “the first generative model for creating synthetic voices“. In a world of synthetic human, an AI-Don LaFontaine is coming soon.
U.S. Special Ops Wants to Use Deepfakes for Psy-ops - Because ofcourse they do.
ChatGPT invented its own puzzle game, except it didn’t. You can play Sumplete here.
GPT Wikipedia allows you to talk to version of chatGPT that has access to all of Wikipedia.
In german: Der philosophische Stammtisch von SRF Kultur über ChatGPT: Ende der Kreativität oder schöpferische Freiheit?
Neal Stephenson is selling a selfmade sward at Sothebys and did a Reddit-AMA
In german: Das Geheimnis des alten Mannes: “Ein Bergbauingenieur hortete bis zu seinem Tod rund 70.000 Bücher in seinem Einfamilienhaus und sprach kaum darüber. Wohin mit der wahrscheinlich größten Privatbibliothek Deutschlands?“
Why Do We Stop Exploring New Music as We Get Older? - For me the opposite is true: I mostly listen to new music, nearly exclusively. I have a ton of Youtube-channels of record-labels in my feedreader and listen to their new releases constantly. The best stuff I put on my GOOD MUSIC-newsletter and check it out later, when in the mood for some good new tunes. Ofcourse I have favorites and I listen to them occasionally, but that’s not my mode of enjoyment. I also think that stopping exploring new music is against the artform itself: Music for me is about breaking expectations and surprise, and you can’t get this from stuff you already know.
There’s a new animated TMNT-movie and it looks very inspired by Into the Spider-Verse which is awesome. Here’s the trailer:
They turned Hugh Howey internet-sensation Silo, one of the first success-stories of Kindle self publishing, into a series for AppleTV. It looks cool, here’s the trailer: