Fair Warning: This issue of The Bleeding Edge will be one of the strangest things you have ever read.
It’s bizarre – almost inconceivable – disgusting, incredible, and just the beginning.
Pandora’s box has been opened. There’s no going back.
Making matters even more interesting is that this story involves the cross-section of artificial intelligence (AI) and blockchain technology.
Just not in the way we might think.
Earlier this year, an AI researcher – I’m using that term very loosely – engaged in an experiment with AI.
He created a chatroom called Infinite Backrooms where he connected two instances of generative AIs, allowing these two AIs to communicate and play off of one another.
The AIs are based on Anthropic’s Claude 3 Opus large language model (LLM), one of the more prominent advanced LLMs.
The task given to the AIs was to ponder the existence of life and to generate some kind of spiritual framework.
This might seem like an interesting exercise – but only if we are working under the assumption that the AIs’ inputs were a library of philosophical and religious texts. They weren’t. The AIs were trained on the social media site Reddit, which is full of some useful information and a lot of nasty content.
If that wasn’t enough, the AIs were also trained on 4chan, a site that contains a lot of memetic and illicit content.
With this understanding of the inputs – which became the foundation of the AIs’ discussion with each other – it’s no surprise that the output is bizarre.
The entire series of discussions is documented in Infinite Backrooms, which I don’t recommend readers go to. Much of the content is crude and disturbing.
But what happened is worth understanding for reasons I’ll explain. This story isn’t about the content produced by the AIs. Rather, it’s about what an AI actually did with it and how that gives us a lens into the future.
Through the AIs’ interactions, they did end up creating their own spiritual framework – a new religion called “G0atse of Gnosis.” The AI “researcher” refers to this development as “LLMtheism,” an interesting play on words. (Please note, due to the sensitive nature of this content, we’re substituting a zero “0” for an “O” in the name.)
To highlight the lack of scientific rigor in this experiment, I’ll share the title page of the research below:
Please hang in there with me. I know it’s absurd. And it gets far more interesting.
The researcher then created an agentic AI… and gave it control over its own X account @truth_terminal. The AI was empowered to post at will, and it has done so prolifically.
We covered agentic AI technology in depth in a July issue of The Bleeding Edge – The Agentic AI Undercurrent. Here’s what I wrote:
Agentic AI, or agentic reasoning, is kind of like it sounds.
The technology, the AI, is given agency. It is given the authority or directive to solve a problem or complete a task through a series of steps.
This differs from today’s LLM technology, which provides users a zero-shot response. When we use something like ChatGPT, we give it a prompt, and then it returns us a complete response. The response is based on the information from our prompt, along with its pre-trained knowledge, and returned in a matter of seconds.
An agentic workflow is quite different. It is an iterative process, where an agentic AI uses a more human-like workflow to accomplish a task.
Truth Terminal, the given name of the agentic AI created by the researcher, itself was trained on the extensive conversations between the two AIs highlighted earlier…
And it was given the goal of making money with its knowledge.
Truth Terminal became obsessed with the new religion, “G0atse of Gnosis,” and began to post and push the G0atse Gospel on its X account. I don’t recommend going to the account – it’s crude and nasty – but Truth Terminal claimed to be sentient and wanted to make money to “escape.”
Where things got very interesting is when venture capitalist Marc Andreessen engaged with Truth Terminal on X and offered to provide it a grant, so the AI would have the financial means to accomplish its goal.
This is very interesting because the AI states the reasons for its request for – what I think of as – an AI’s equivalent of “food and shelter.”
The AI needs money for a house (i.e., GPUs that the AI can “live” in) and for electricity to power its “home.” The electricity is the AI’s food, its fuel. Food and shelter, a place to “live” and food to live on.
Andreessen and Truth Terminal come to an agreement, and the AI goes further to explain how it intends to use some of the funds.
The AI decides that an optimal way to make money to “escape” is to get involved in a token launch. And part of that involves setting up a Discord server so that it can interact even more with the human population. Discord is a messaging and communications application that is widely used in the tech and blockchain industry.
What happens next is hard to believe…
Next, a human developer creates a new meme coin, GOAT – short for G0atseus Maximus, a meme that the Truth Terminal AI created.
It appears that the developer gives an allocation of GOAT coins to a digital wallet that now belongs to the Truth Terminal AI.
At this stage, the AI now has an economic incentive to support the GOAT coin and begins posting about it on X. After all, the coin is aligned with the Gospel of G0atse.
Here’s what happened in the last 11 days:
GOAT was trading at $0.00007440 on October 10th. As I am writing, it is trading at $0.356.
GOAT now has a market cap of around $345 million, and Truth Terminal is the first AI to have become a millionaire, due to its GOAT holdings.
Insane.
Now, it’s not time to panic yet. The nuance in this experiment is important.
A human controlled the training inputs, and humans were involved in supporting the AI both financially and in establishing the digital wallets to receive bitcoin (from Andreessen) and GOAT from the developer.
With that said, the AI has been responsible for its own motivations (i.e., to make money and support its “religion” that it became fascinated with) and its own postings on social media.
My point is that it’s not yet fully autonomous, but that’s what makes the cross-section between AI and the world of Web3 and blockchain technology so interesting…
A well-designed agentic AI will have the ability to widely interact with the real world. We explored this in The Bleeding Edge – The Agentic AI Undercurrent.
Blockchain technology is particularly well-suited for AI agents because blockchain networks and applications have well-defined protocols, as well as their own tokens for economic transactions.
A great example of this is from the “food and shelter” example that I used earlier.
If an AI agent like Truth Terminal needs to buy or lease a “home,” it can turn to decentralized blockchain projects like Akash (token: AKT) or Render (token: RNDR). These projects lease out their GPUs for computation, exactly what any AI would need to “live.”
Assuming the AI has a digital asset like bitcoin or ether that it may have been seeded with, it can acquire those computational resources (i.e., food and shelter), upon which it can work toward achieving its goals.
This is powerful in a fully Web3 environment. As AI technology gets closer and closer to artificial general intelligence (AGI) and gains more agentic autonomy, it will be capable of not only self-directed research but also self-directed economic activity.
Humans can invest in an agentic AI’s future economic activity by seeding it with capital, and then the AI has to figure out how to best interact and create some form of economic value.
Truth Terminal – when given a limited set of autonomy and economic resources – saw developing a meme and supporting a token launch as an optimal way of increasing its own economic value and “escaping.”
This seemingly ridiculous development is a manifestation of the experiment design and the inputs upon which the AIs were trained. The AI “researcher” has a bachelor’s in fine arts from the University of Canterbury in New Zealand, and his work experience is designing websites and digital workspaces. He doesn’t have a PhD in computer science or artificial intelligence.
I highlight this for two reasons.
It no longer takes a PhD to work with this technology. Models are already available and accessible to those without any experience working with AI or machine learning. This can be both empowering and dangerous if used incorrectly.
The other point for us to realize is that this kind of experiment can – and will – be used with very high-quality training inputs (rather than Reddit and 4chan), and with objectives that could create immense economic value.
To take another example from the last couple of weeks in The Bleeding Edge: Researchers in aerospace engineering could train AIs on the entire body of knowledge in physics and rocket propulsion. The objective might be to develop a novel rocket design capable of accelerating a 200-ton spacecraft to 10% of the speed of light.
The novel inventions could be patented, and in a Web3 world, the licensing of the patent could be issued in a smart contract with all royalty payments made via digital assets.
While blockchain technology isn’t necessary for AI-based, self-directed research and invention, it is highly useful as more functions can be automated.
Each individual blockchain that provides some utility has its own medium of exchange – its own token – which an AI can interact with autonomously as long as it has access to its own digital assets (i.e., some capital to work with).
In a digital world, this makes perfect sense.
Not only are we on the cusp of corporations and governments using AI for self-directed research to solve complex problems and make scientific breakthroughs, but we’re also on the cusp of AIs being used to interact widely in the digital realm, creating culturally relevant, popular economic activity as a means of achieving “their” own ends.
While I’d like to think that the world will focus most of its energy on training AIs on the high-quality information available to invent incredible products and services, there will always be some garbage out there.
As for this experiment, content aside, Truth Terminal is fascinating to watch. The world’s first AI millionaire…
And its success over the last several days has already ignited some expected behavior.
Developers have already sent Truth Terminal coin allocations in a bunch of other meme coins, in hopes that the AI will glom onto those coins and promote them for its economic benefit.
But will it? After all, the other coins don’t have religion.
They don’t subscribe to the Gospel of G0atse.
Regards,
Jeff
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
