ABCDE: Looking at AI+Crypto from the perspective of the primary market

Source: ABCDE, Medium

More than a year after the release of ChatGPT, the discussion about AI+Crpyo has become lively again in the market recently. AI is regarded as the most important game in the bull market of 24-25 years. One of the reasons is that even Buterin himself published "The promise and challenges of crypto + AI applications" (Crypto+AI Application Prospects and Challenges) to discuss the possible exploration directions of AI+Cryto in the future.

This article will not make too many subjective predictions, but simply from the perspective of the primary market, give a rough overview of the entrepreneurial projects that combine AI and Crypto observed in the past year, and take a look at the entrepreneurial projects Specifically, from what angle did the researcher enter the market, what achievements have it made so far, and what areas are still being explored.

1. The cycle of AI+Crypto

Throughout 23 years, we have talked about dozens of AI+Crypto projects, among which we can see obvious cycles.

Before the release of ChatGPT at the end of 22, there were very few AI-related blockchain projects in the secondary market. The main ones that everyone can think of are FET, AGIX and other established projects. The AI ​​that can be seen in the primary market Not much is relevant either.

January to May of 2023 can be said to be the first concentrated period of outbreak of AI projects. After all, Chatgpt has had a great impact on people. Many old projects in the secondary market have pivoted to the AI ​​track. In the primary market, AI+Crypto projects are also discussed almost every week. Similarly, the AI ​​projects during this period feel relatively simple. Many of them are "copycat" + "chain modification" projects based on ChatGPT. There are almost no core technical barriers. Our In-House development team often spends A project basic framework can be reproduced in just one or two days. This also led to us talking about a lot of AI projects during this period, but in the end nothing was done.

The secondary market began to turn bearish from May to October. Interestingly, the number of AI projects in the primary market also dropped sharply during this period, and it was not until the last month or two that the number became active again. , discussions, articles, etc. about AI+Crypto on the market are also enriched. We have once again entered the "grand scene" where we can meet AI projects every week. After half a year, it is obvious that the emerging batch of AI projects have a clear understanding of the AI ​​track, the implementation of business scenarios, and the combination of AI + Crypto have significantly improved compared to the first batch of AI Hype periods. Although the technical barriers are still not strong , but the overall maturity has reached a higher level. It was only in our 24th year that we finally made our first bet on the AI+Crpyto track.

2. The track of AI+Crypto

Victoria gave a prediction from several relatively abstract dimensions and perspectives in the article Prospects and Challenges:

We summarize these AI projects currently seen in the primary market from a more specific and direct perspective. Most AI+Crypto projects are centered around the core of Crypto, which is "technical (or political) decentralization + commercial assetization."

There is nothing to say about decentralization, what about Web3... According to the category of assetization, it can be roughly divided into three main tracks:

• Capitalization of computing power

• Capitalization of models

• Capitalization of data

• Capitalization of computing power

This is a relatively intensive competition Dao, because in addition to various new projects, there are also Pivot of many old projects, such as Akash on Cosmos and Nosana on Solana. Moreover, the tokens have skyrocketed after Pivot, which also reflects the market’s interest in AI competitions. Tao is optimistic that although RNDR focuses on decentralized rendering, it can actually serve AI. Therefore, many classifications have also classified all computing power-related assets such as RNDR into the AI ​​track

Computing power assets can be further subdivided into two directions according to the use of computing power:

• A One is "decentralized computing power used for AI training" represented by Gensyn;

• One is "decentralized computing power represented by most Pivot and new projects" Use it for AI reasoning";

You can see a very interesting phenomenon in this track, or we are not optimistic about the contempt chain:

• Traditional AI → decentralized reasoning → decentralized training

• < p>Those who come from traditional AI majors are not optimistic about decentralized AI training or reasoning

• Those who come from decentralized reasoning are not optimistic about decentralized training

The reason is mainly technical, because AI training (especially large model AI) involves massive amounts of data, and what is more exaggerated than the data demand is the bandwidth demand caused by high-speed communication of these data. In the current Transformer large model environment, training these large models requires a large number of 4090-level high-end graphics cards/H100 professional AI graphics cards purchased computing power matrix + 100G-level communication channels composed of NVLink and professional optical fiber switches. You To say that this thing can be implemented in a decentralized manner, hmm...

AI reasoning requires far less computing power and communication bandwidth than AI training. The possibility of decentralized implementation is naturally much greater than that of training. This is also Why are most computing power-related projects focused on inference, and the only people trained are big players like Gensyn and Together, which have raised over 100 million yuan? But equally, from the perspective of cost performance and reliability, at least at this stage, centralized computing power is still far better than decentralized reasoning.

It is not difficult to explain why when looking at decentralized reasoning and looking at decentralized training, they think “you can’t do it at all”, while looking at decentralized training and reasoning for traditional AI, they feel “the training technology is unrealistic” , "The reasoning is commercially unreliable."

Some people said that when BTC/ETH first came out, everyone also said that the distributed node calculation model was not reliable compared to cloud computing. Didn’t it work in the end? Then it depends on the future needs of AI training and AI reasoning for the dimensions of correctness, non-tamperability, and redundancy. Simply focusing on performance, reliability, and price cannot be better than centralization for the time being.

Capitalization of models

This is also a track where projects gather together, and it is also a track that is easier to understand than the capitalization of computing power, because ChatGPT is the most famous after it became popular. One of the applications is Character.AI. You can consult with sages such as Socrates and Confucius, chat with celebrities such as Musk and Ultraman Sam, and even fall in love with virtual idols such as Hatsune Miku and Raiden General. All of this, These are the charms of large language models. The concept of AI Agent is deeply rooted in people's hearts through Character.AI.

What if Confucius, Musk, and General Thunder are all NFTs?

Isn’t this AI X Crypto? !

So it is not so much the assetization of the model, but the assetization of the Agent based on the large model. After all, the large model itself cannot be put on the chain. It is more about the Agent mapping based on the model. Turn it into NFT to create a visual sense of AI X Crypto that is similar to “model assetization”.

Now there are Agents in the circle who can teach you English, and there are also Agents who can fall in love with you. There are all kinds of agents, including Agent search and Market Place and other derivative projects.

The common problem in this track is that there are no technical barriers. Basically, it is the NFTization of Character.AI. Our technical masters at In-House used existing open source tools and frameworks to create one overnight. He talks like BMAN and sounds like BMAN's Agent. Second, the degree of integration with the blockchain is very light, a bit like Gamefi NFT on ETH. Essentially, the Metadata may only be a URL or a hash, the model/Agent is on the cloud server, and the transactions on the chain are only It's just a matter of ownership.

The assetization of models/Agents will still be one of the most important tracks of AI More Native projects may appear in the future.

Data capitalization

Data capitalization is logically the most suitable for AI+Crypto, because most traditional AI training can only use what is visible on the Internet. Data, or to be more precise - public domain traffic data, these data may only account for less than 10-20%. More data is actually in private domain traffic (including personal data), if these traffic data can be used To train or Fine-Tune large models, we can definitely have more professional Agents/Bots in various vertical fields.

What is the slogan that Web3 is best at, Read, Write, Own!

So through AI+Crypto, under the guidance of decentralized incentives, we can release the data of personal and selfish flow, capitalize it, and provide better and richer "rations" for large models. It sounds like This is a very logical approach, and there are indeed several teams working intensively in this field.

However, the biggest difficulty in this track is that data is difficult to standardize like computing power. Decentralized computing power The model of your graphics card can be directly converted into how much computing power, but the quantity, quality, use and other dimensions of private data are difficult to measure. If decentralized computing power is ERC20, then decentralized computing power The assetization of AI training data is a bit like ERC721, and it is still a project of PunkAzuki, with many Traits mixed together. The difficulty of liquidity and market is not even a little bit more difficult than ERC20, so currently doing AI data Capitalization projects are a bit difficult to move forward.

Another thing worth mentioning in the data track is decentralized annotation. Data capitalization is used in the "data collection" step, and the collected data needs to be processed before feeding it to AI. Processing, this is the step of data annotation. This step is currently mostly centralized labor-intensive labor. Use decentralized token rewards to make this Labor Work decentralized, label it to Earn, or disperse the work in a similar way to a crowdsourcing platform. , is also an idea. We have seen a small number of teams currently working in this field.

3. The missing puzzle pieces of AI+Crypto

Let’s briefly talk about the missing puzzle pieces of this track from our perspective.

The first is technical barriers. As mentioned before, the vast majority of AI+Crypto projects have almost no barriers compared to traditional AI projects in Web2. They rely more on economic models and token incentives to put effort into front-end experience, market and operations. Of course, this It is understandable that decentralization and value distribution are the strengths of Web3. However, the lack of core barriers will inevitably lead to a sense of X to Earn. I still look forward to more teams like RNDR, the parent company of OTOY, with core technologies to show their talents in Crypto.

The second is the current situation of practitioners. As far as what has been observed so far, some teams of entrepreneurs in the AI ​​X Crypto track know AI very well, but do not have a deep understanding of Web3. Some teams are very Crypto Native, but have limited attainments in the AI ​​field. This is very similar to the early Gamefi track. Either they know games well and think about the Web2 game chain modification, or they know Web3 well and think about the innovation and optimization of various gold farming models. Matr1x is the first team we have met on the Gamefi track that has a double-A understanding of games and Crypto. This is why I have written before that Matr1x is one of the three projects that I have "decided on as soon as we talked about it" in 23 years. We are looking forward to it. You can see a team that understands double A in the fields of AI and Crypto in 2024.

The third is the business scene. AI The combination of AI and Crypto among various projects currently seen on the market is somewhat "rigid" or "rough" and does not bring out the optimal competitiveness or composability of AI or Crypto. This is also inconsistent with The second point mentioned above is closely related. For example, our In House R&D team has thought of and designed a better combination method. Unfortunately, after looking at so many projects on the AI ​​track, we still haven’t seen any team entering this segment, so we can only continue to wait.

What, you ask why we, a VC, can think of certain scenarios before entrepreneurs on the market? Because there are 7 great people in our In House AI team, 5 of whom have PHD background in AI. As for the ABCDE team’s understanding of Crypto, you know...

The last thing I want to say is that although from the perspective of the primary market, AI x Crypto is still very early and immature. But this does not prevent us from being optimistic that AI X Crypto will become one of the main tracks of this bull market in 24-25 years. After all, AI liberates productivity and blockchain liberates production relations. Is there any better way to combine the two? :)