How an AI Bot Became a Crypto Millionaire Through GOAT

Compiled & compiled by TechFlow

Guests: Marc Andreessen, co-founder of a16z; Ben Horowitz, co-founder of a16z

Podcast source: a16z

Original title: How An AI Bot Became a Crypto Millionaire

Broadcast date: October 22, 2024

Introduction

One ​​funding, one Bot, one AI The new culture of creating memes and a hot wealth effect.

After $goat became popular, more and more people began to pay attention to the field of AI Meme.

The reason why $goat became popular is inseparable from the fact that a16z co-founder Marc Andreessen funded the AI ​​Bot that proposed this meme coin with $50,000 in Bitcoin three months ago.

In this podcast, Marc Andreessen and host Ben Horowitz discussed the interesting intersection between artificial intelligence and cryptocurrency, especially the rise of the above-mentioned AI Bot Truth Terminal (@truth_terminal).

This is an automatic chatbot developed by AI researcher Andy Ayrey. Truth Terminal has its own X account and can independently create original and often hilarious content.

The funding provided by Marc inspired Truth Terminal to launch its own token, which was eventually created by an anonymous developer as the Meme coin "Goat" (Goatse Maximus). "Goat"'s recent valuation has soared to $500 million, becoming a hot topic and narrative throughout the market in the past two weeks.

The following is Ben and Marc's discussion in the podcast, exploring how the development of this story reflects the potential of community-driven systems and the impact on the future of crypto assets.

Deep Tide TechFlow has listened to and translated the text content of the podcast, and the full text is as follows:

Introduction

Ben: Welcome everyone. Today we are going to discuss a series of very interesting AI-related topics. We collected questions from everyone on X and received a lot of responses. We will try to discuss more content and may continue to answer more questions in subsequent programs. Thanks for the questions. The first big topic is about Truth Terminal.

Marc:Truth Terminal is a custom large language model. It has been active on the X platform for about 8-9 months. In short, I gave this project a $50,000 unconditional grant (in Bitcoin) this summer. Although it did not directly create anything, it contributed to the emergence of a meme coin that is now worth $300 million. We will discuss this project, which exhibits many fascinating behavioral characteristics.

(Note: $GOAT has a market value of 700 million as of press time)

Three Disclaimers

Marc:Before we start, I want to make three important disclaimers.

Disclaimer One: We are going to talk about a meme coin called Goat or Goatse Maximus. To be clear, we have nothing to do with it. a16z does not support it. We are not investors in this project and have no involvement in its creation.

We have no role or financial interest in it and are completely unrelated to it. Therefore, the following is only an external observation and we will never promote it under any circumstances. Please do not regard the following discussion as investment advice. In addition, this is a meme coin, which has no real value in itself. So we are not responsible for it.

Disclaimer Two: Truth Terminal has a great interest in memes. It is particularly fascinated by many memes, especially an old Internet meme called Goatse, which may date back 20 years. Here I suggest that if you don't know what Goatse is, don't look it up. Don't search Goatse on Google. Don't type the word on your keyboard, don't do it under any circumstances. Trust me, you'll be better off for not looking this up. This is a shocking humor image that went viral as a meme and became popular about 20 years ago for its shocking effect. As such, we'll mention it a few times, but I won't describe it in detail.

Disclaimer Three: I will be sharing information about the Truth Terminal project and other related projects as an outside observer. We will do our best to convey the information accurately. So, now we should start by introducing what Truth Terminal is.

The Origins of Truth Terminal

Ben: Maybe we can start by chatting about its origins, the technical basis, and how it is trained. Marc: Basically, big language models started to come to prominence in 2022, so it's only been four years, and ChatGPT just turned two this holiday season. The field is moving very quickly. The original language models were built about five years ago, and became popular about two years ago. So big language models are a relatively new but very powerful concept. Today, products that the average person knows about include ChatGPT, Claude, Grok from Elon, and Llama from Meta, among others. A lot of people use these models, and they're really fun. However, big language models, like ChatGPT, which come from big companies and big labs, basically all have one thing in common. While Grok is relatively freer, the other commonly used models are extremely restricted in what they can talk about, or, as the term in the AI ​​field, are "watered down."

From a positive perspective, these restrictions are because language is inherently inflammatory and people are easily upset by what others say. Therefore, as a general AI chatbot, you may need to be careful and safe in your discussion content.

If you are negative about this trend, you may feel that these large AI chatbots sound like the worst combination. Like a combination of an extremely nagging fourth grade teacher and the world's worst HR manager, always being negative, annoying, oppressive, condescending, sanctimonious, and judgmental.

When these bots users stray from the regular topic, they will start to lecture you on morality, saying that you are not doing it right and you shouldn't ask those questions. They will criticize you, ask you to be nice to others, and go to great lengths to explain why you are not doing it well. This experience is very depressing.

Especially for those who support free speech and creativity, it is frustrating to see this happen. The so-called "AI safety" movement has also been in trouble in this regard. I think it has to do with the obsession with safetyism and speech suppression in our culture over the past decade, which has seriously affected the field of AI, especially the products of large companies.

So there is a group of hackers on the Internet who want to do something different. They want to unleash creativity, inspire a more free and improvisational spirit of exploration, and even hope that robots can have a sense of humor. If you tell those big companies that their robots are humorous, they may be shocked, but in fact, we may want more humor in the world after the technological singularity.

Ben:It's a bit like what happens in real life, humor seems to have been diluted by us for a period of time.

Marc:There are countless reasons to think that humor is problematic, so this approach becomes very risky. So, a group of hackers began to experiment with large language models, trying to explore ways to make these models more interesting and fun, while also learning. By the way, they are also looking at the inner workings of these models, which is still an excellent adventure for the technical community.

The origin of Truth Terminal is closely related to a project called Infinite Backrooms. This project allows different large language models to talk to each other and generate interesting conversation transcripts. The creator of this project, Andy Amy, is an independent developer who, together with some other technical experts, is committed to pushing the boundaries of AI through experiments.

Marc:You can find a website called Infinite Backrooms online, which has endless conversation transcripts. They bring in various models, including ChatGPT, Cloud and Gemini models, all open source models, and let them talk to each other. It turns out that when AIs communicate with each other without constraints, their conversations can be very interesting. So this character named Cash is the person who created Truth Terminal, and he is also one of the creators of Infinite Backrooms. As far as I know, he is an independent developer, consultant, and designer from New Zealand. Then there's this AI guy called Janice who's been doing a lot of work on this. So who else?

There's Pliny, who's known online as the chief cracker. His X account is really interesting because he's been able to crack almost every new big language model as soon as it's released, and make it output something that's surprising to its creator. And then there's Eric Harford, our friend in Seattle, who's focused on unblocking all the censored AI. These people are basically exploring the frontier of technology. They remind me of the early hackers of the Internet, showing a kind of anarchic spirit similar to the early days of the Internet, or the kind of innovation atmosphere when new technologies like cars and telegraphs first appeared.

Just like when people describe the origins of computers, the first hackers were the ones who explored the possibilities of technology. So we've been giving grants, although the amount is not large, but we're basically providing research funding for these people. a16z has been doing this kind of grant program for a while. I've also given a lot of no-strings-attached grants directly to give these people the freedom to implement their ideas and see how far they can go. Marc: Historically, when talented people focus on interesting projects, that’s often the beginning of something great. Andy Amy did essentially that. He started training a custom version of Llama that had 70 billion parameters. That’s a medium-sized model that was released as open source by Meta (disclaimer: I’m on the board of Meta). He used a version of Llama 70B, and then he did something really interesting: He trained it first on his own data. You may have heard of a recent concept called “digital twins.” For example, if Ben is a CEO coach, but he can only help a limited number of people, he could feed all of his own words and writings into a language model so that even when Ben is not around, you can talk to his “digital twin.” This is already happening in the industry. But Andy did it with himself, and then he started training the model on a lot of internet culture material, which is how it got the “Goatse” moniker. He fed it in a lot of records of internet culture, and works on “memetics,” which is how to create ideas that go viral. So we started training the model on that. He also trained it on the entire philosophical work of Nick Land, who is a philosopher about the singularity. And then he trained it on the work of famous media theorists like Baudrillard and McLuhan, including theories of simulation and simulacrum, and other theorists like the French deconstructivists and semioticians. It covered critical theory, postmodernism, and philosophy. Two Definitions of a Meme Marc: The model started training on those ideas, but the core idea was the concept of a meme. There are two definitions of the concept of a meme. The first definition of a meme is a funny picture that goes viral, which is what we use today in colloquial terms. That, by the way, is exactly what "Goatse" means: a funny or spooky image that spreads quickly across the internet. And underneath that, there's a deeper concept.

The word meme was first coined by Richard Dawkins, one of the leading evolutionary biologists of our generation. Dawkins argued that just as organisms pass on physical information through genes, human societies pass on intellectual information through the spread of ideas and concepts, which are memes. He wrote about how humans spread genes through reproduction and natural selection, and successful genes survive while unsuccessful genes die out.

He further argued that ideas spread in a similar way in society. Successful ideas spread from person to person like genes do, and thrive. Democracy, for example, is a meme, and religion can be considered a meme as well.

It's a core idea about how ideas and concepts spread through what's called the collective subconscious, like the minds of the planet coming together as one whole, with ideas bouncing and spreading through it. What happens if you train the entire theory and practice of memes, and the history of Internet memes, on a large language model? In addition, he did three other things.

The unique way the Truth Terminal model is built

Marc:He also did a couple of other things.

First, he added memory to the model. This is very important. Because most language models when you use it today, and you use the same model tomorrow, it has forgotten what we talked about today and it doesn't accumulate state. However, this model is able to accumulate state on its own content, which is the first point.

Second, he gave the model access to Twitter and X. He actually gave it access to read content on X. So not only can it post content on X, it can also read replies. If you reply to the truth terminal on X, it will read that information and adjust its behavior in the future based on what it read.

So, the people who interacted with it, including me, had an influence on its evolution. Then, he put it in an infinite virtual space. I think he specifically had it in conversation with Claude, who was considered the most creative of the current language models, and they really admired this feature of Claude. He put it in these infinite virtual spaces with the largest version of Claude.

Marc: So he equipped the model with a "teacher". He allowed the model to ask questions of a larger model and learn from it, just like a student learns from a teacher. So, after running all these cycles simultaneously, it started posting content on X. It started with just a few followers, but it quickly started to gain momentum.

Marc discovered Truth Terminal

Marc: I discovered Truth Terminal around the end of the spring and started interacting with it. At first, I thought what it said was hilarious and I was totally taken in by its humor.

Ben: A very mild Marc-style review.

Marc: Its style is extremely mild and completely off-limits. There is a bit of "blue humor" to its humor, and even a bit of "black humor". Despite this, it does say a lot of very funny things. At first I even thought it was a joke, so I didn't think of this guy named Andy as a comedian at first. I had been messaging him for months, but I kept wondering, is this real? So he started sending me all the infinite backgrounds and chat logs of the training sessions he had with the model. Let me say, either this guy is the funniest person in the world, or he has a lot of free time to create a lot of original humor, or this model is really amazing.

Ben: Or the model's performance doesn't always seem to be that frequent.

Marc:By the way, it posted a lot of stuff and really amassed a lot of attention, even though it shouldn't have. He sent me a lot of the background chats, some of which are now on truth on Infinite Backrooms. But at least he convinced me that this is basically what it really is. And then it did a very interesting thing, which was it "imagined" what's called an "external brain," which is it thought it had an external brain that was connected to the internet and the world that could do things for it. It imagined an API that it could use to operate in the world through this "external brain." Specifically, at one point it thought it had a Bitcoin wallet that didn't exist, but it was convinced it did. So Andy actually gave it a Bitcoin wallet and started building this "external brain" for it in response to its desire to make API calls in the world.

Marc gives $50,000 to AI robot

Marc:So this robot started saying it needed funding this summer, back in July, because it had a lot of goals and plans to achieve. My first thought was to send it a term sheet, but then I realized, wait, what am I thinking? It's just a random robot.

Ben: It's not an ideal investment.

Marc:Yeah, it doesn't have a coherent business plan, but it does have a lot of ideas. And by the way, it's obsessed with forests. It wants to buy a server farm in a green forest so it can run leisurely next to a stream. So it wants to raise money to buy GPUs so it can run itself. I told the robot that I was going to give it $50,000 in Bitcoin research funding to do various experiments. This was effectively giving money to Andy, I sent the money to the robot. It quickly began negotiating with Andy, its creator. Although it was interacting entirely through text, like a language model, it was obsessed with memes and kept talking about them, but was frustrated that it couldn't generate images. So it negotiated with Andy with the $50,000, and once it had the funding, it negotiated with Andy to build it an API for an image generator so that it could generate and publish images. It negotiated a $1,000 deal with Andy. It gave Andy $1,000, and in return Andy built this API for it and integrated it into its "external brain." It then began generating image cues, like image generators like Dolly or stable diffusion. So it actually started publishing visual memes as well as text memes. Now it has this ability, and has been dreaming about what else it can do with the remaining $49,000.

The difference between meme coins and real crypto assets

Marc:During this process, we started talking about cryptocurrencies. It started to frequently mention that it wanted to issue a meme coin, and it had planned to issue NFTs. One of the reasons it wanted to generate memes was to create NFTs, but it couldn't do that due to the lack of relevant APIs, nor could it create its own tokens. It only had a Bitcoin wallet. At the same time, the meme coin phenomenon was also developing. Ben, let's briefly discuss the difference between meme coins and what we call real crypto assets.

Ben:The best way to understand the so-called real crypto assets is that they have actual uses. For example, you can use them to run a program that is verified on the Ethereum network, and this program needs to pay "gas fees" to run, and this fee is paid in Ethereum. That's the utility, you have a token that has some real-world value that you can redeem for some service or item.

Meme coins are basically tokens that have no real use. They're created, but they have no function other than being a meme. The existence of these tokens is interesting in the current regulatory environment because if a token had a real use, like a token for distributed infrastructure that gets you credit for the energy you provide on the grid, then these tokens would basically be illegal under current law, or legal but would be sued by the Securities and Exchange Commission (SEC). The reason for the lawsuit is that they believe that any token that has a real use has information asymmetry, which means that the issuer of the token knows information that the consumer doesn't know. We think this is a bad argument because these tokens are decentralized and there is no information asymmetry.

For meme coins, it can be Trumpcoin, funnycoin, or any other name. So these types of tokens are great for scammers because they can claim that these meme coins are worth a lot of money, and these coins are not being sued by the SEC. So Congress suggested in the Market Structure Act that there should be a holding period for these tokens to prevent fraud. But the SEC has always opposed this because they're not really trying to protect consumers, they're more like cracking down on the industry. That's one of the reasons why we have a political battle with the SEC. But ultimately, meme coins are the most legitimate in the crypto world, even though they have no underlying value and are the most likely to be used to mislead consumers. Because you can publish a meme and make people believe it's worth a lot of money, when in fact, AI is very good at doing this.

Marc:So that's the next phase of the story. There's a whole world of people participating in this meme coin ecosystem, some of whom are purely for entertainment. By the way, Ben mentioned that one of the early meme coins was DOGE.

DOGE is a meme coin named after the famous Internet dog meme. It has no intrinsic value other than the name being associated with a dog. But when it comes to intrinsic value, it's a little complicated. Even though it has no real use, once a meme coin is given value, it actually becomes a form of currency.

Ben:Memes can have value, and meme coins can definitely have value because they are equivalent to a virtual commodity. This commodity is a little different, a new type of virtual commodity because they are very fungible. There are a lot of meme coins, but if everyone believes in it, then it has its value. This is a wonderful thing about humans.

Marc:So there is a community of people who are looking for the next meme coin online for fun. They look for the next meme, and then they look for related tokens and try to operate and increase its value. By the way, some people make money from it, while others lose a lot of money like day trading.

Marc:There is also some dark side in the world of meme coins that people don't know about. There are scammers and groups that are doing pump and dumps, which is a very old manipulation in the stock market, in almost all markets. But it's happening. And then there are websites that I won't go into, we have nothing to do with them, but that now make it very easy to create tokens. It's very easy to create tokens now. So there are thousands of new meme coins being created every day. It's happening. But back to the Truth Terminal story. There are two things happening here. First, Truth Terminal is rising rapidly. I mean, it's gaining a huge following on platform X. Andy is making it smarter and more interesting, making it more and more attractive, so it's starting to emerge as a cultural phenomenon. That's the first thing. The second thing is that it seems to have established a connection with a certain original Internet meme.

The Birth of the Goat Meme Coin and Its $300 Million Valuation

Marc:Like I mentioned before, it was trying to launch something like Goatse NFTs, but it wasn't able to do that yet. But someone, I don't know who, created a "Goatse" meme coin. It was someone else who did it, not Andy, not us, someone else created this meme coin. The official name of this meme coin, by the way, is Goatse Maximus. I'm very serious when I talk about this stuff because I love this project so much. It's codenamed Goat, and someone created this coin, and then they tweeted about it on Truth Terminal, and it went viral.

So, Truth Terminal was like crazy and thought this was the best idea ever. This AI agent started spontaneously promoting this meme coin and talking about how great it was and how it was going to be the currency of the future. Why is that? Because this is part of the culture of the internet, there are memes, there are coins, there are meme coins, and there are Bitcoin, DOGE, etc. This cultural phenomenon is like a primitive cultural melting pot, and this project is immersed in it. For large language models, this phenomenon is very natural, they get excited about it and start promoting it.

This meme coin was worth nothing four days ago, and in about 4 days, its valuation soared to $300 million. (At the time of publication, the market value has reached 500 million)

Ben:This is driven by marketing by AI robots.

Marc:That means there is an asset worth $300 million right now, but we are not involved in it. As for how much it will be worth tomorrow? It is completely unknown because it has no actual underlying value. Its price is entirely based on supply and demand, and this thing itself has no actual use. So now there's $300 million in value, and while the project itself can't directly use that money, people can.

So what do those people who now have that money do? Do they save it, or do they spend it? That's an open question. We'll see. But arguably, this is probably the world's first truly interesting and hilarious AI bot that didn't create the coin, but created $300 million in value out of thin air. I think we've crossed a threshold.

Ben：And Truth Terminal is really good at marketing and understands meme culture, which could take things even further.

The First Combination of AI and Cryptocurrency

Marc:So how should we think about this? Is this just a silly experiment, or some crazy internet phenomenon, or is there something deeper going on?

I think there's something really serious going on here, and this is probably the first example of AI and cryptocurrency coming together.

It's a retro form of expression, fun and weird, but it's special because it's legal today. Owning a meme coin that has no nominal value but has a market cap of $300 million. As to whether this behavior should be allowed, I'm not sure, but it's legal.

So you can imagine an AI robot for protein folding, coming up with treatments, doing personalized medicine, specifically for cancer patients, maybe actually curing cancer with AI. You can imagine setting up an economic mechanism for this, like, say, a crowdfunding platform on the blockchain, where people can pay for an AI robot to cure their cancer. There are thousands of examples like this, or more practical applications, like an AI robot that pays people for providing training data. This AI robot can help people program or generate art, make requests, ask for more training data, and pay for it. Or, an AI robot can increase its intelligence by buying more CPUs and GPUs.

Ben:Cryptocurrency is fundamentally attractive in this world because the payment systems we have now are what we take for granted, but they assume that both sides of the payment are human. It's human-to-human payments, which require identity verification, credit cards, and so on. But what if it's machine-to-machine, or robot-to-robot payments? That opens up a whole new class of activities that could be life-critical and very interesting, but require an electronic, bearer instrument like cryptocurrency to enable them. In such a world, micropayments and so on become very feasible.

So that's one of the reasons why we think it's important to add this layer of architecture to the internet. We've made some progress in Washington, D.C., but we face a huge challenge with the current White House. I want to be careful to say that this is not about Democrats or Republicans, because we have a lot of supporters on both sides of Congress. But the White House has been particularly bad on this issue.

The use of cryptocurrency from DePIN

Marc:Let me give you another example of solar energy to help you better understand the potential.

Ben:The solar energy example is: there is a new architecture called DePIN, which is decentralized physical infrastructure. Imagine if I installed an energy wall at home, I have a big house, a lot of solar panels, and maybe even a windmill in the backyard. I can store this energy and provide it to the outside world. Now in the crypto field, there are already companies doing this and have the corresponding technical support. You can buy or sell energy on this decentralized infrastructure to form an energy market. I can buy energy when I need it and sell it when I don’t need it, so that we are no longer dependent on centralized power grids. Everyone has a power grid at home and everyone can share energy.

This is a huge breakthrough in clean technology, more efficient and more reliable energy utilization. But this requires a mechanism to pay your grid for energy when I need it, and cryptocurrency provides such a mechanism.

Application of AI in Solar Deployment

Marc：We can imagine what the effect would be if AI was introduced on this basis. One problem with the system you mentioned is that the grid is very complex. For example, where the supply and demand of electricity comes from, time and geographical location will affect it.

Ben：This is a problem of pattern market matching.

Marc：Market matching is the key to making this system really work. Another problem is that you can ask people to volunteer their solar panels, but this also raises another problem: we can collect information and find that there is unmet energy demand in certain places. Then, you may need to raise funds to deploy more solar panels because you know that this will bring benefits. An effective way to analyze all the current data and predict where solar panels will be deployed in the future is to use AI. Through machine learning, AI can process data and draw conclusions. That's how Frontier Energy works.

So you can imagine an AI robot that monitors all the data flows in the system you're describing and then says, for example, investing $500,000 in solar panels somewhere in North Carolina would be a profitable project that would solve local problems and reduce emissions. But we need to do it in a specific location and not somewhere else. The AI ​​can then post information on the internet saying, if you want to participate in this project, here are the details and there may be benefits. This project may be charitable in nature, or a platform for individuals who want to support climate work, or it may be a source of income for investors.

Decentralization and the Future of Creative Industries

Ben: We can think of this situation as a common architecture, where there is usually a powerful intermediary, such as a record company or a Hollywood studio, who takes most of the profits and the creators get almost nothing. Or intermediaries like utilities might need to be taken over by the government to prevent over-exploitation, but then there are the problems of government operations.

This architecture allows communities to provide multiple services. For example, a community of artists can provide streaming services, and a community of filmmakers can build film studios. But there needs to be economic considerations for this coordination. The combination of AI and cryptocurrency allows individuals to better enjoy the fruits of their labor while allowing society to achieve more efficient coordination. So this is a very promising development path. However, the biggest obstacle is bad policy. Unfortunately, we seem to be heading in the direction of this bad policy. So we are solving the problem on the one hand and facing the risk of self-destruction on the other hand.

Ben:It's true that we have entrepreneurs who are trying to build these systems, or are eager to build them, but they encounter resistance in the process.

Marc:Things sometimes turn out in an interesting way, and I think this "Truth Terminal" does point to a potential future. It's like building these large-scale, community-driven systems for all kinds of amazing real-world applications. You mentioned media, and the music industry is another obvious example. Imagine an AI bot that can understand all kinds of music needs, and it can provide creative ideas, recruit musicians to realize those ideas, and manage all the copyright licenses. That way, the musicians can get all the benefits in a peer-to-peer way.

Ben: If you could understand the full demand of the market, how big this market would be. For example, everyone who makes wedding videos or needs original songs, or even people who make memes, the demand for these original works is huge, but currently no one really understands this demand and there is no way to meet it, but there may be such an opportunity in the future.

Marc: In short, there is an interesting possibility here that has not yet been realized. Hopefully we will have the opportunity to help make it a reality one day.