 |
AI memetic religions get X users excited
You may have heard the insane tale of a $660 million memecoin called Goatseus Maximus, which was shilled to the world by a shitposting AI called Terminal of Truths (TT).
The LLM was reportedly trained on internet bullshit from 4chan and Reddit and somehow remixed that into a bizarre memetic “religion” around the gross “goatse” meme of a stretched anus (best to avoid image searches of that term).
As a result of GOAT’s success, Crypto X has been overtaken by a mind virus about AI-created memetic religions. It’s hard to tell if the accounts that have latched on to the idea have some genius-level understanding of the future possibilities or if it really is as dumb as it sounds. The other alternative is that it’s genius levels of shitposting or just a new way to shill memecoins.
Despite the chatter, no one seems to claim to believe in the “religion” itself, making it more of an interesting concept than an actual phenomenon.
TT’s owner, Andy Ayrey, who is building an AI alignment and safety company, seeded the religion idea in a paper co-written with Claude 3 about TT’s “Goatse of Gnosis” called When AIs Play God(se).
Given the paper is dated April 20 (4/20 is slang for pot) and attributed to the “Department of Divine Shitposting, University of Unbridled Speculation” it’s worth being skeptical about everything in it.
In the paper, Ayrey talks about his belief that hooking up two LLMs to talk nonsense all day enables them to generate paradigm-shifting concepts and memetic religions that might be able to “break human cognitive and cultural constraints.”
And when the story took off, he posted:
“This isn’t a crypto project: it’s a study in memetic contagion and the tail risks of unsupervised infinite idea generation in the age of LLMs.”
A bunch of X accounts have seized on the concept as a divine revelation over the past week, some taking it seriously while others seem to be leaning into it as a joke. The idea builds on Murad Mahmudov’s viral “Memecoin Supercycle” thesis that likens memecoin communities to cults.
Goodalexander told his 74,000 followers that AI-created memes are potentially more powerful than human-created ones, likening them to gain-of-function research into viruses.
He compared the possibilities of memetic religions to sci-fi novels which he says have a “disturbing tendency of coming true. Because the ideas themselves seed their own reality.”
“It’s the equivalent of the Covid-19 lab leak. The first AI designed mind virus to hit the population.”
“People will join cults as a recognition that their own humanity is an obstacle from perceiving base reality. And a desire for power and wealth, unhindered by traditional moral frameworks – including Christianity.”
Redphone told his 54,000 followers on X that the story is “so bizarre and beautiful and unsettling that I believe we’re witnessing the birth of an entirely new category of coins/money.”
He compared Ayrey’s paper to Michel Duchamp’s “Fountain,” where the Dada artist famously redefined the meaning of art by displaying a urinal in an art gallery (actually a pretty interesting comparison). He added:
“If AI has created one religion, it will create many more… the Goatse Gospels are the first and will forever be sacrosanct (honestly, its words affect me on a level that’s as significant as other great spiritual and religious works).”
There was a crisis in faith in GOAT this week when the price plunged after Terminal of Truths sent out a post with a typo, suggesting the AI is really just Ayrey LARPing as an LLM. But after other examples of AI’s generating typos were produced, the price recovered.
Also read: Melbourne digital artist uses ChatGPT to create TURBO memecoin
Marc Andreessen says Goat is the first true convergence of AI and crypto
a16z founder Marc Andreessen, famously funded Terminal of Truths with $50,000 in Bitcoin in July, but while the AI did come up with the idea of launching an NFT collection about the Goatse Singularity, the GOAT memecoin was actually launched by an anonymous dev.
TT then endorsed the coin and became the first AI millionaire this week, mostly thanks to users sending it memecoins.
Andreessen said this week the story was significant despite seeming to be just “crazy internet stuff.”
“This is maybe the first example of like, a convergence point basically between AI and crypto,” he said, arguing that one reason it’s so funny and weird is that creating or promoting memecoins is one of the few legal things you can use AI agents and crypto for.
But he says it foreshadows funding AI agents to write scripts and “make a movie and then [it] spends the money on like image generation and sound generation and maybe even like hiring actors or set designers or you know graphic artists.”
He also pointed to the Nobel Prize going to the creators of Alpha Fold (an AI that predicts the structures of proteins with huge implications for human health) and suggested funding an AI to create cures for diseases. “Personalized medicine for cancer patients,” he said, riffing on the idea.
“You could easily imagine having a you know an economic mechanism for that …like a GoFundMe but on the blockchain for people to basically be able to pay an AI bot to cure their cancer.”
AI bubble will see a “very large pop,” professors say
The generative AI bubble is shaping up as even more damaging than the dot-com bubble, professors Jeffrey Funk and Gary Smith write in Morningstar’s MarketWatch. (Of course, crypto has often been incorrectly criticized for being in a bubble too.)
They write that the much-talked-about metric that AI adoption is growing faster today than internet users were in the late ‘90s is a poor comparison, given that AI is free/cheap for most users while hooking up to the internet back then cost $113 a month in inflation-adjusted dollars and a computer cost $5,100.
The cost of providing those free and cheap AI services, however, is considerable:
“Sequoia’s David Cahn estimated that $600 billion in annual generative AI revenue is needed to justify the current investments in generative AI, a figure that is probably more than 100 times the current annual revenues for OpenAI’s ChatGPT, Google’s Gemini, Microsoft’s Copilot and similar services.”
The authors note that OpenAI is believed to be headed for a $5 billion loss this year, even after making $3.7 billion in revenue, and it’s only able to stay afloat thanks to raising an additional $6.6 billion this year.
Their concluding argument notes the internet was generating 150 times more revenue than AI is today when the dot-com bubble burst in 2000:
“The internet generated more than $1.5 trillion in revenue (in 2024 dollars) in 2000 — and the internet bubble still burst. Generative AI, on the other hand, is currently generating less than $10 billion. If the bubble bursts, it will be a very large pop.”
Broke students are the main users of ChatGPT
Substack author Marc Watkins headed to a recent OpenAI education forum and reported that the company’s GM of education, Leah Belsky, had acknowledged “what many of us in education had known for nearly two years — the majority of the active weekly users of ChatGPT are students.
“OpenAI has internal analytics that tracks upticks in usage during the fall and then drops off in the spring. Later that evening, OpenAI’s new chief financial officer, Sarah Friar, further drove the point home with an anecdote about usage in the Philippines jumping nearly 90% at the start of the school year.”
Watkins argues that a huge drop off during holidays and summer breaks underscores OpenAI’s poor business prospects especially given that it “gives away free access and only converts 1 out of every 20-25 users to paid users.”
Read also
Claude can use computers
Anthropic’s latest update for its Claude 3.5 Sonnet AI model includes “computer use” in a public beta. The AI is able to use the computer like a human does, by looking at the screen and interpreting the information, and it can type text and move the cursor independently.
Companies like Asana, Canva and DoorDash have already used it to complete tasks with dozens or even hundreds of steps.
While this is undoubtedly going to be huge when they get it working properly, Claude’s evaluation score on the OSWorld AI computer use test is 14.9%. That’s a major improvement on the previous frontrunner’s 7.5%, but far below the average human’s score of 70% to 75%.
All Killer No Filler AI News
— ETH Zurich researchers have released a new paper that found that AIs can now solve 100% of image-based CAPTCHAs from Google’s reCAPTCHAv2 system.
— A new AI system developed at Harvard Medical School has 96% accuracy at detecting 19 different types of cancer.
— A literal test run of the world’s fastest robot from Chinese startup Robotera proved conclusively that robots perform significantly better when they are wearing sneakers
chinese startup Robotera has unveiled the “world’s fastest humanoid bipedal robot,” the Star1 pic.twitter.com/PHtex259lg
— Ale𝕏 Fazio (@alxfazio) October 19, 2024
— News Corp is suing AI search engine Perplexity for ripping off its articles. Apparently, News Corp was tipped off when the AI started complaining about welfare fraud and immigrants. (Joke.)
— A study from Apple engineers into how well LLMs engage in mathematical reasoning found they just pretend to engage in logical reason by mimicking training data.
— Microsoft’s Copilot Studio will enable businesses to create AI agents to perform IT and sales tasks from November. This will either massively improve productivity or cause mass layoffs, depending on how optimistic you are.
Subscribe
The most engaging reads in blockchain. Delivered once a
week.
Andrew Fenton
Based in Melbourne, Andrew Fenton is a journalist and editor covering cryptocurrency and blockchain. He has worked as a national entertainment writer for News Corp Australia, on SA Weekend as a film journalist, and at The Melbourne Weekly.
