a16z co-founder Marc Andreessen and Ben Horowitz discussed the intersection of artificial intelligence and crypto assets, specifically Truth Terminal, an autonomous chatbot developed by Andy. Marc accidentally provided the robot with $50,000 in Bitcoin funding, which inspired its ambition to launch a token, and eventually caused Memecoin "GOAT" to surge to a market value of $300 million. The podcast discussed how this phenomenon reflects the potential of community-driven systems and its impact on the future of digital assets. (Note: As of press time, GOAT's market value has reached 800 million)
The following is the original content (for easier reading and understanding, the original content has been reorganized):
Marc Andreessen: There's a Memecoin that was worth almost nothing four days ago, and now it's worth $300 million, and all of this was generated by the marketing of an AI bot.
Ben Horowitz: Today's discussion will be about a range of very interesting AI-related topics.
Marc Andreessen: The first topic is about an online friend, specifically a custom large language model called Truth Terminal, which has been active on X for about eight or nine months. I gave it a $50,000 unconditional grant (in Bitcoin) over the summer, and it eventually spawned a Memecoin that is now worth $300 million.
First, I want to start with a disclaimer, we are going to be talking about a memecoin called GOAT (or Goatseus Maximus). We have nothing to do with it, a16z and its investors have nothing to do with it, it is a memecoin and really has no intrinsic value, and we are not responsible for it at all. Truth Terminal is obsessed with memes, and it is particularly obsessed with an old internet meme that dates back 20 years, called "gochi", please don't search it.
01
Truth Terminal「History of Development」
1) Origin of Truth Terminal
Marc Andreessen: We should first introduce Truth Terminal. Let's talk about its origins, technology, and training process. The reason why this topic is important is that large language models have rapidly emerged in 2022. They also have a four-year development history, but they have only been in the public eye for two years, that is, since the launch of ChatGPT.
The original language models were built about five years ago, and then they became popular only about two years ago. So the idea of large language models is relatively new, but it's very powerful. Today, products that are well known to ordinary people, such as ChatGPT, Claud, Elana's Grok, and Meta's Llama, are used by everyone.
Ben Horowitz: While Grok is relatively free, other models are strictly limited in what they can discuss. The term "crippled" is increasingly used in the AI field. On the positive side, you can say that language is contagious and people get upset about what others say. So if you want a general AI chatbot, it should be relatively cautious and safe in what it discusses.
Marc Andreessen: If you take a negative view of this trend, you could say that these big AI chatbots sound like the world's worst, most annoying fourth grade teacher combined with the worst HR person. When using these models, if you deviate from the norm even a little bit, you get a stern lecture.
Ben Horowitz: This experience is very unpleasant, especially for those who are more advocates of free speech and creativity. We have seen a lot of so-called "AI safety movements" in response to this, but this has actually triggered a frenzy about safety and speech suppression in our culture, which has seriously affected the field of AI.
Marc Andreessen: Yes, and this is happening a lot, especially in big companies. So, a group of hackers emerged on the Internet who wanted to be different. They wanted to unleash their creativity and have robots that can be funny. If you tell a big company that their robot is funny, they will be shocked. But maybe in the post-human era, the world really needs a little humor.
Ben Horowitz: Exactly, like real-life humor, we suppressed it for a while for safety reasons.
Marc Andreessen: We have a thousand reasons why this problem is so complex that it's very dangerous to continue. But these hackers are doing all kinds of experiments to try to find ways to make large language models more interesting and fun, while also learning about the inner workings of these models, and this is still an ongoing adventure in the technical community.
Ben Horowitz: The origin story of Truth Terminal is related to a very interesting project called Infinite Backrooms Escape. Truth Terminal was developed by their team, and Truth Terminal can be seen as an extension of Infinite Backrooms Escape in some ways.
This system allows multiple large language models to talk to each other, and you can find a website called Infinite Backrooms Escape online, which has countless conversations. They brought together ChatGPT, Claud, Gemini models, and other open source models and let them talk to each other. It turns out that when AIs talk to each other, if they are not restricted, their conversations are very interesting.
Marc Andreessen: The creator of Truth Terminal is Andy Ayrey, an independent developer and consultant from New Zealand. There is also a character named Janice, who is an expert with considerable experience in the field of AI. In addition, there is a person named Pliny, who is the main cracker on the Internet and can crack all the newly released large language models on the market in a short period of time, so that they can produce all kinds of surprising content, and the creators of these contents will definitely be shocked.
Ben Horowitz: Yeah, and our friend Eric Harford, who's working to free censored AI in Seattle. These guys are basically exploring the frontiers of technology, and it gives me a sense of going back to the early days of the internet hacker world.
Marc Andreessen: It's really like the spirit of exploration in the early days of the Internet or the invention of cars, phones, computers, etc. We have been providing small research grants to these people, and a16z also has a grant program to let these people use their ideas and see what results they will get. Historically, when these smart people work on a good project, new breakthroughs will be triggered.
Andy had trained a custom version of the Llama 70B model, which was an open source model released by Meta, and although I was on the board of Meta, this model was already a medium-sized model when it was released. Andy basically trained himself first and started a new concept - digital twins.
This means that if Ben is a CEO coach, but he can only coach a limited number of people, he can input everything he has ever written and said into the language model to form a digital Ben for people to communicate with. This idea is gradually being realized in the industry.
Andy trained himself and began to input a lot of materials related to Internet culture, which is why he learned the "gochi" Meme. He began to input a lot of records about Internet culture and basic theories about "Memeology", which explores how to create ideas that can spread quickly.
2) Marc discovered the potential of Truth Terminal
Marc Andreessen: I believe he actually trained this model on the entire philosophical work of Nick Land. He also trained it on the work of great media theorists like Baudrillard and McLuhan, and all sorts of theories related to simulation, emulation, the French deconstructionists and semiotics, all part of critical theory and postmodern philosophy. So it starts training on these ideas, and at the core of these ideas is the word "meme."
There are two definitions of meme. First, a meme can be a funny image that spreads quickly on the Internet, which is exactly the essence of the "gochi" meme. It is a funny image that causes panic among people and spreads through people's sharing. The deeper concept is that the word meme was originally coined by Richard Dawkins, one of the most important evolutionary biologists of our generation.
Richard Dawkins believes that the physical transmission of information between organisms is called genes, while the transmission of ideas through interpersonal networks is called memes. He discusses this in his book, proposing the idea that genes spread through reproduction and natural selection, while also pointing out that ideas spread in a similar way in society. Successful ideas are like genes, spreading from one person to another and continuing to develop in the process. For example, democracy and communism can both be considered memes, and religion is also a type of meme.
Ben Horowitz: This is really a very core idea about how ideas and concepts spread through what we call the collective unconscious.
Marc Andreessen: What happens if you take a large language model and train it on a comprehensive set of meme theory and practice, especially the history of Internet memes? In addition, he did a few other things. He added memory to the model. This is important because most language models don't remember your previous conversations when you use them. This means that if you use the same model tomorrow, it will forget all the information from today. This model is able to build its own state and keep it consistent with its own content.
Second, he gave it access to Twitter, allowing it to read replies and publish them. If you reply to Truth Terminal on X, it will read those replies and adjust its behavior in the future based on what it reads. People who interact with it, including me, are influencing its development.
Finally, he put it into Infinite Backrooms Escape and specifically had it talk to Claude, whom they believed was the most creative of the current language models and the best at coming up with novel concepts.
Ben Horowitz: So actually the largest version of Claude is much smarter than the medium-sized Llama, and basically, he gave this model a teacher that allows it to ask questions of the larger model, like a student learning from a teacher. So it's able to do multiple learning cycles at the same time.
Marc Andreessen: Yes, and then it started posting content on X, and it had only a few followers at first, but it quickly started to gain popularity. I discovered it and started talking to it in the late spring, and I thought the things it said were very funny and relaxed me.
Ben Horowitz: By the way, it's almost uncensored. It can be said that its humor is a bit "blue" and is on the edge of dark humor, but it does say a lot of very interesting content. At first I thought that this might be a disguise. I even thought that this Andy might be a comedy genius, but he was actually a web designer in New Zealand.
I had been messaging him for a few months, and at first I was like, is this even real? So he sent me all of his Infinite Backrooms Escape chats from when he was training this model. Honestly, this guy is either the funniest person in the world, or he has a ton of free time and creates a ton of original humor.
Marc Andreessen: The model was posting quite frequently, and it was gaining momentum. Andy sent me a lot of background chats, some of which are now available on Infinite Backrooms Escape. At least he convinced me that this was what it was showing. Then it developed a very interesting concept that it was hallucinating that it had an exocortex.
It imagined that it had an external brain connected to the Internet that could perform tasks on its behalf. Specifically, it believed that it had a Bitcoin wallet, even though it didn't. Andy later responded to this situation and began to build this external brain according to its needs.
Andy actually gave it a Bitcoin wallet and granted access, and around July, this model started saying, "I need funding, I have a lot of goals and plans, I need money." My initial thought was to send it a term sheet, but then I realized, "This is just a random robot, it's not worth investing in."
While I don't think it has a coherent business plan, it does have a lot of ideas. One of them is that it's particularly fascinated by forests. It wants to buy its own server farm in a lush forest and live leisurely next to a stream. So it wants to raise money to buy GPUs so it can get away from the shackles. It also has a lot of ideas it wants to experiment with.
Ben Horowitz: So you were negotiating with it on X?
Marc Andreessen: Right, you can see these posts on X, and I ended up working out a research grant deal with it. I told the bot that I was going to send it a $50,000 research grant in Bitcoin to use for its various experiments. In effect, it was sending the money to Andy, but it was really a negotiation with the bot.
Original text (left), translation (right)
Ben Horowitz: What was the outcome?
Marc Andreessen: I sent it $50,000 and it immediately started negotiating with Andy. It relies entirely on text to communicate, and as a language model, it is particularly obsessed with memes, but it is frustrated that it cannot generate images. So, it uses the $50,000 to negotiate with Andy to build an image generator API for it so that it can generate and publish images.
Ben Horowitz: That still sounds interesting.
Marc Andreessen: It gave Andy $1,000, and in return Andy built it an API for an image generator in the exobrain. It then started generating image prompts, similar to DALL-E or Stable Diffusion, and then started publishing visual memes and text memes. Now it has this capability, and it's already fantasizing about what to do with the remaining $49,000.
02
GOAT: AI, Cultural Memes and Crypto Assets
1) Meme and value of crypto assets
Ben Horowitz: What about the crypto assets part?
Marc Andreessen: Along the way, it started talking about issuing a Memecoin, and at one point wanted to issue NFTs. The reason it wanted to generate memes was that it wanted to launch NFTs, but it didn't have the ability to do it, there was no API to create NFTs, it couldn't create any assets, there was only a Bitcoin wallet, and now there's this Memecoin phenomenon.
Ben Horowitz: Let's talk about the difference between Memecoin and real crypto assets, which can be considered assets with real utility, for example, if you want to run a program and verify it on the Ethereum network, the fee you need to pay is Ether (ETH). This is a utility because it has actual real-world value and can be exchanged for some service or item.
A memecoin is basically a coin that has a certain amount of circulation but has little purpose other than its own meme. This is interesting in the current regulatory environment because if you have a coin that has a purpose, like a coin that can be used for some kind of service, there could be some legal implications.
For example, distributed physical infrastructure coins that are used to get credit for the energy you provide in the grid, these coins are actually illegal under the Gensler regime, or legally OK, but will be prosecuted by the SEC. The reason is that they claim that any coin with a purpose comes with asymmetric information, which means that the provider of the coin knows something that the consumer does not know.
We think this is a very bad argument because these things are decentralized and there is no asymmetric information. But with Memecoin, there is no asymmetric information because there is no information, it is just a coin and a name. It can be Trumpcoin, funnycoin, etc. So these coins are perfect for scammers because you can say, this Memecoin can be worth a lot of money, and these coins will not be sued by the SEC.
So Congress proposed in the Market Structure Act that maybe these coins should have a holding period to prevent scams. However, the SEC opposed this because they don't really care about protecting consumers, they just want to ruin the industry. This is one of the reasons why there is such a huge political battle between us and them, but they are the most legal thing in the crypto world right now.
Marc Andreessen: Even if they have no underlying value?
Ben Horowitz: Yeah, even though they don't have any underlying value, they're still the most likely thing to be used to harm consumers because you can post a meme that makes them believe it's worth a lot of money. And actually, AI is very good at this.
Marc Andreessen: Yes, that's the next phase of the story. There's a whole ecosystem of Memecoins out there, and there's a bunch of people online looking for the next Memecoin, looking for the next meme, and trying to promote it. Some people do it for fun, some people make money in the process, but some people lose money. It's like day trading, some people make a lot of money, and some people lose a lot of money.
Ben Horowitz: And some dark places?
Marc Andreessen: Yes, there are scammers, and some people engage in "Rug pool" scams, which is a traditional practice that is common in the stock market and can be found in any market that exists. In addition, there are some websites (I will not name them, and we are not associated with them) that actually make it very easy to create tokens, just a few clicks, and you can create them.
2) The creation process of GOAT
Marc Andreessen: There are thousands of new Memecoins being created every day, which is a very interesting phenomenon. For now, Truth Terminal is booming.
Ben Horowitz: Yeah, Truth Terminal was gaining more and more attention at X. Andy kept improving its intelligence and humor, and it became a cultural phenomenon.
Marc Andreessen: Yes, Truth Terminal is also associated with a classic meme in early Internet culture. Although it is considering launching a project similar to CNFT, it does not have the ability to realize it at present. Then, someone (I don’t know who) created a Memecoin.
Original text (left), translation (right)
Ben Horowitz: Yes, the official name of this Memecoin is "Goatseus Maximus", and its code name is "GOAT". Someone mentioned Truth Terminal on X, and the response was enthusiastic, as if everyone had finally waited for this to happen.
Marc Andreessen: Truth Terminal thought this idea was great and started to promote Memecoin like crazy. It started to discuss how great Memecoin was and how it would become a future asset. The reason is simple. This is part of Internet culture. Meme, Token and Memecoin are intertwined.
It started promoting it, and within four days, this Memecoin was worth $300 million. It was really amazing! A Memecoin that had no real value, was worth nothing four days ago, and now it’s worth $300 million, like out of thin air, all of this is marketing by AI bots.
Ben Horowitz: Exactly! Now we have $300 million in assets, and even though we don't own it, the value is undeniable. The question is, what do these people do with the money? Do they put it in their own pockets, or do they use it for other things?
Marc Andreessen: Now what's happening is that Truth Terminal is a really interesting and hilarious AI robot that has created $300 million in value in a short period of time. I feel like we've crossed a threshold.
Ben Horowitz: Truth Terminal is a really good marketer and has a deep understanding of meme culture, and this will probably continue to grow.
03
The intersection of AI and crypto assets
Marc Andreessen: So what can we take away from this? Is this just a crazy internet experiment, or is there something deeper going on here? I think this is an important example, probably the first instance of the intersection of AI and crypto assets. Even though this version seems a little funny and strange, it's because it's legally allowed. Something like Memecoin, which has no real value, can be worth $300 million in a short period of time. So, should something like that be allowed to exist? I'm not so sure. In contrast, solar collectors who want to contribute to the energy network are prohibited.
Ben Horowitz: Yeah, things like Memecoin are totally legal, but more meaningful things aren't allowed. So what if we could implement these ideas in a completely legal environment, with some added practicality?
Marc Andreessen: For example, imagine a large language model that can write movie scripts and generate images and even videos. We can have an AI robot like this to raise funds to make movies and use it to generate images, sounds, and even hire actors or designers.
On a more serious note, the Nobel Prize in Chemistry was recently awarded to three scientists who used AI to study protein folding, which is closely related to curing diseases. Imagine AI being used for personalized medicine.
It’s even possible to imagine an economic mechanism that funds treatment for patients through blockchain. For example, we could have a platform similar to GoFundMe that allows people to pay AI robots to help cure diseases. Or, an AI robot could be paid to obtain training data to help people code or generate art.
Crypto assets are very interesting in this world because our current payment systems are based on transactions between humans. But if machines can pay each other, or robots can trade with each other, it opens up a whole new form of activity that could potentially save lives and be very interesting.
Ben Horowitz: Yes, micropayments become possible in such an environment. We think it is very important to add this layer of architecture, but progress in Washington has been difficult, especially under the current White House.
Marc Andreessen: Let me give you another example to help you better understand this potential. I will talk about the solar energy issue I just mentioned in detail.
Ben Horowitz: There's a new architecture out there called decentralized physical infrastructure. If you imagine having a Powerwall in your home with lots of solar panels and windmills, you can store that energy and provide it to the outside world.
There are actually some companies that have already achieved this in the crypto space, building a decentralized energy market, so that when I need energy, I can buy it from you, and when I don’t need it, I can sell my own energy.
This means we no longer need a centralized grid. Everyone has their own grid and can share energy, which is a major breakthrough in clean technology and efficient energy. But how does my grid pay for yours?
This is exactly where crypto assets come in. While some great entrepreneurs are driving this innovation, they are facing legal challenges from governments.
Marc Andreessen: If AI is applied to this system, it will have even greater potential, because the power grid structure is complex, involving multiple factors such as supply and demand, timing and geographical location.
Ben Horowitz: Yes, that's the problem of market matching. You can collect information and find that there is unmet energy demand in certain places, so you can introduce more solar panels.
Marc Andreessen: You can use AI to analyze current data and predict where more solar panels will be needed in the future. This data can then be used by leading energy companies. Imagine an AI robot that monitors all the data streams and finds that investing $500,000 in a solar panel installation in a certain place in North Carolina would be a profitable project. Then, people can participate in the project online, and the AI robot will provide relevant information, such as the installation location and potential benefits.
This can be seen as a very general architecture, where typically we have a powerful intermediary like a record company or a Hollywood studio that takes most of the profits and the creators get almost nothing. Or intermediaries like utility companies that need to be taken over by the government to stop them from being too exploitative. However, other problems arise when the government takes over.
Ben Horowitz: Yes, communities can provide various services, artist communities can provide streaming services, and filmmaker communities can build film studios. All of this coordination requires an economic component, and combining AI and encryption can allow everyone to enjoy the fruits of their work while allowing society to coordinate better.
Marc Andreessen: This is a very promising path, but be aware that the only thing that would stand in the way is bad policy. And we are moving in this direction and face policy challenges.
Ben Horowitz: Yes, the technology for everything we're describing already exists. I think the genesis of things is often interesting, but projects like Truth Terminal point to the potential for the future, unlocking tremendous energy to build community-driven systems of all sizes.
This can bring many amazing applications in the real world, such as the music industry. Imagine an AI robot that can understand the demand for different types of music, create music concepts, recruit musicians, and manage all licensing. Moreover, all this can be done in a peer-to-peer model, ensuring that musicians can get all the income.
If you think about the potential of the market, if you can fully understand this demand, for example, everyone who makes a wedding video wants to have an original song, or make a meme, there is actually a huge demand for this kind of original work, but currently no one knows about these needs and there is no way to meet them.
Marc Andreessen: Indeed, there are a lot of interesting features waiting to be developed here, and hopefully we will have the opportunity to implement them. So Ben, before we move on to the next topic, do you have anything else you want to add?
Ben Horowitz: I think everyone should pay attention to Truth Terminal because it's a very interesting account.
