Discussion on the Crypto AI Track

Host: Alex, Research Partner at Mint Ventures

Guest: Max, YouTube channel curator; Lydia, researcher at Particle Network

Understanding Crypto AI

Alex: Today we are going to talk about the highly关注的 Crypto AI track. The first topic is, what do both of you think about the Crypto AI track? In your opinion, what business problems is the Crypto AI track trying to solve? What is the urgency of these problems?

Max: I think the emergence of Crypto AI is to solve two main problems. The first is from a humanistic perspective, centralized AI itself has some issues that need to be addressed, such as censorship. Crypto AI can solve these problems through decentralization. Another interesting point is the introduction of incentive mechanisms. The main representative of Crypto is Token, and with Token, all decentralized AI can leverage this incentive mechanism to try out more different approaches. For example, Bittensor creates different subnets through the Token mechanism, with each subnet responsible for researching different things. This method connects the open-source code. Open source has always been something everyone wants to pursue, but the biggest problem AI researchers face when doing open source is that there isn't a way to reward the progress of open source. By connecting to Crypto and Token, there is now a way to reward them for continuing to research open source, rather than each company privatizing its research results. Overall, what Crypto AI is doing or can do is to reward open-source models, promote openness, and encourage the development of decentralization through the incentive mechanisms of Crypto and Token.

Lydia: From a business perspective, I feel that the answer is not particularly clear to me, especially at the Crypto level. While there is a saying that "AI can improve efficiency, and Crypto ensures fairness," upon closer consideration, from the current stage and the perspective of business value, the urgency to improve efficiency is clearly greater than ensuring fairness. I always think of the article Alex wrote earlier about the underlying value of Web3, which mentioned that the underlying value of Web3 is broader freedom and cheaper trust. Therefore, excellent Web3 projects need to identify the shortcomings of traditional services in freedom and trust, and then provide a more competitive solution. In the context of Crypto AI, does AI need greater freedom? From a technical implementation perspective, computational resources and data supply are limited, so the freedom of AI is also limited. From an ethical standpoint, a truly free AI is something we can hardly imagine. Is the trust cost of AI too high now? I don't think so. Although many people mention the data black box issue, more attention to this problem comes from experts, scholars, or media practitioners, rather than ordinary users. On the other hand, if we try to solve it through on-chain methods, the cost currently seems to be higher.

I believe the greatest value of Crypto AI at present may not be directly reflected in alternative solutions on the current commercial level, but more on the narrative level. It opens up people's imagination, allowing the seemingly unrelated but particularly cutting-edge and stylish technologies of Crypto and AI to collide in everyone’s minds. We need to give these two technologies time; perhaps the problems they are best suited to solve belong to the future, rather than the present.

From the very beginning, I felt that it was a long-term exogenous narrative. Speaking of long-term, it is because AI, especially consumer-level AI, has had a tremendous impact on our real world. It truly is a disruptive transformation. Not only can we see it from the data, such as ChatGPT breaking one million users in just a few days and surpassing one hundred million monthly active users in two months, but we can also directly observe the frequency with which people around us use AI. From the perspective of the capital market, OpenAI has a valuation in the hundreds of billions, and Nvidia is valued at trillions. Every time they hold a press conference, it basically dominates the headlines of major media outlets. This transformation is happening too rapidly and too fundamentally, so AI will not be a fleeting trend; it will certainly be a long-term narrative and may even become one of the most important sources of philosophical discussion in the coming century.

At the same time, it is also external. Crypto and AI actually have no relationship after their inception, and there is even a competitive relationship on the talent level. During the Crypto bear market from 2022 to 2023, AI's appeal in this area overshadowed Crypto. It wasn't until this year that we began to tell the story of mutual empowerment between the two. Ultimately, compared to crypto-native narratives like DeFi and NFTs or transformed narratives like GameFi, AI is an external narrative. We can also see that, like earlier today, the asset prices of AI narratives such as Worldcoin, Render, and Near fluctuate entirely based on the conditions of the AI industry. Therefore, I believe that the long-term external narrative is my initial understanding of Crypto AI, and I still hold this view.

Max: I want to add something. You said that AI is an external thing that already existed in Web2, and we initially thought that Crypto and AI were two unrelated things, but then they suddenly came together. However, I believe that from another perspective, Crypto AI is the only strong demand from Crypto for AI after the DeFi Summer in 2020. For instance, in GameFi, we add the incentive mechanisms of Crypto into games, but Crypto is just an embellishment for GameFi. Today, if GameFi were to leave Crypto, people wouldn't play the game just because the Crypto incentive mechanism is great; they would play it because the game is fun. DeFi, on the other hand, is another level; it is a hard demand. I think Crypto AI is the second strong demand that can follow after DeFi, having seen so many narratives.

With the advancement and use of AI, we are bound to discover some centralized issues, only we haven't identified them yet. Unlike the financial systems of the financial world that have existed for possibly 100 or 200 years, we recognized problems in the financial system only after the 2008 financial crisis, which made us realize that there are issues that need to be addressed in this system. That's why everyone feels that DeFi is what we need. I believe Crypto AI is in a similar position. It's just that users' exposure to and familiarity with AI is not as extensive as that of the financial system, so we haven't yet seen people truly feel that "I really need this Crypto AI thing."

When it comes to why Crypto is a hard requirement in the narrative of Crypto AI, it is because many things need to incorporate incentive mechanisms to be realized. Like you just mentioned about wanting to become more efficient, I believe there are already some specific projects that can achieve this. For example, Decentralized Compute has been in operation for a while. When you compare decentralized computing power with centralized computing power, you will find that as long as some performance bottlenecks are overcome, decentralized computing power is basically the primary demand. You wouldn't want to use centralized computing power, you wouldn't want to use products like AWS or Microsoft Azure because they are too expensive or for other reasons. I genuinely believe that for Crypto AI to break out and continue to develop, it must be more efficient, better, and cheaper than traditional products, it has to be that way. People won't just want to use Crypto AI to "support decentralization"; it must be better than the original products. This is what Crypto AI needs to accomplish now. We can slowly see this prototype emerging, but we cannot expect Meta to release a free 3.5 billion parameter LLM model every time. We need to find a way to continuously build this. I think this is a direction that requires ongoing effort.

Classification of Projects in the Crypto AI Track

Alex: Crypto AI is a relatively large field, and within it, there are many different types of projects with various business models addressing different problems. Based on your understanding of the Crypto AI field, if you were to categorize the projects within these fields, what kind of logic would you use to classify them?

Lydia: A very common classification method is Crypto empowering AI or AI empowering Crypto, which are two major ideas. Currently, we see more AI empowering Crypto, meaning Crypto projects are trying to add some AI attributes. Previously, it might have involved integrating APIs, creating a Web3 version of a chatbot that can answer questions about the project, or using AI to improve the code of Web3 projects, or having AI participate in the formulation of profit strategies. Now, it’s more about AI agents issuing tokens, which has little to do with the efficiency improvements and fairness that AI can bring; it’s more about projects wanting a new narrative.

The second approach is that if Crypto empowers AI, the ceiling is indeed higher, but it is more difficult to implement and validate, requiring more time. The monument in the direction of Crypto empowering AI states that Crypto can penetrate into the AI technology stack, enhancing its privacy and transparency, but the landing cycle may be a bit longer. So currently, it is more about finding an opportunity for Crypto to improve a certain link in the AI industry, such as producing GPUs, which can focus on how Crypto can aggregate and incentivize idle computing resources, reduce costs, and then move on to creating data markets and algorithm markets. They all want to find product market fit from the perspective of freedom. However, I believe that the demand in this area is not particularly easy to prove at this stage. If you look at the GPU usage data from iOasis, you will find that the proportion of individual users is still relatively small. The total rental income from GPUs for individual users may be around 1,000 dollars per day.

I think the current breakthrough point, or the exception, might be that Coinbase and Base are moving in this direction, which is AI agents combined with payments. Of course, the payment aspect is just an added bonus, so the prerequisite is that the AI agent must be good enough and useful enough. This is my two classification methods.

Max: I mainly divide it into three different tracks. The three tracks are the architecture layer, resource layer, and application layer. The architecture layer is more like a foundational structure where you can develop different AI projects on top of this base architecture, allowing various resource layer projects or application layer projects to be built on this architecture layer. If you have a good understanding of blockchain, you might imagine it as a layer 1 blockchain, among other infrastructure, referred to as the architecture layer. Projects like Bittensor, Near, and Sahara are all considered part of the architecture layer.

After the architecture layer is built, there will be a resource layer on top of it. This layer consists of various resources needed for AI development, such as computing power, data, models, etc., and is referred to as the resource layer. Some example projects like Akash or Render provide decentralized computing power, while projects like Vana can provide decentralized data, which are called resource layers.

On top of the resource layer and architecture layer, the layer that is closer to consumers and user usage is called the application layer. I place AI agents here, which are more aligned with the actual needs of users, for example, to accelerate your usage of DeFi. So these are the three main tracks. Since the Crypto AI narrative has just emerged, people are still unsure how to classify it properly, and there is no consensus method. However, this framework structure seems to be a categorization method that can resonate with the current Crypto track.

Opportunities and Challenges of Crypto AI

Alex: What do you think is the main challenge facing Crypto AI now? Besides challenges, what kind of industry or narrative opportunities do you think there will be for Crypto AI in the next one to two years?

Max: I think the main challenge is that Crypto AI is still too early. Most projects have seen their market values rise significantly, like Bittensor which has reached a market cap of 5 billion dollars. Behind such market caps, there is likely more speculation. I believe we really need to find Product Market Fit or discover some truly usable applications, and there are still relatively few applications that can develop these. If we look at these applications, I even feel that some of them are still quite early.