ABCDE: Viewing AI+Crypto from the perspective of primary market

Source: ABCDE, Medium

After more than a year of release of ChatGPT, the market’s discussion on AI+Crpyo has become lively again recently. AI is regarded as one of the most important tracks in the bull market in 24-25 years. Even V God himself posted “The promise andchallenges of crypto + AI applications” (Crypto+AI application prospects and challenges) explores the possible exploration directions of AI+Cryto in the future.

This article will not make too many subjective predictions, but will simply summarize the entrepreneurial projects that combine AI and Crypto observed in the past year from the perspective of the primary market to see which angles entrepreneurs are starting from.What achievements have been made in the market and what are the areas still being explored.

1. AI+Crypto cycle

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

Before the release of ChatGPT at the end of 2022, there were very few blockchain projects related to AI in the secondary market. The main ones you can think of are several old-fashioned projects such as FET and AGIX. There are also not many AI-related projects that can be seen in the primary market.

January-May 23 can be said to be the first concentrated outbreak period of AI projects. After all, Chatgpt has brought too much impact. Many old projects in the secondary market have Pivot gone to the AI ​​track, and the primary market is almost every week.All of them can talk about AI+Crypto projects.Similarly, AI projects during this period feel relatively simple. Many of them are “skin-skin-filled” + “chain modification” projects based on ChatGPT, with almost no technical core barriers. Our In-House development team often spendsA project basic framework can be reproduced in one or two days.This also led to a lot of AI projects we talked about during this period, but in the end we didn’t take any action.

The secondary market began to turn bearish from May to October. It is interesting that AI projects in the primary market have also dropped a lot during this period. It was not until the number of people became active again in the past one or two months. Regarding AI+Crypto on the marketDiscussions, articles, etc. are also enriched.We are once again entering the “glorious scenery” of AI projects that we can meet every week.After half a year, I clearly felt that a number of newly emerging AI projects have understood the AI ​​track, and the implementation of business scenarios, and the combination of AI+Crypto has been significantly improved compared to the first batch of AI Hype period. Although the technical barriers are still not strong., but the overall maturity has taken a step forward.We have also entered 24 years before we finally made our first bet on the AI+Crpyto track.

2. AI+Crypto track

In the article “Foreground and Challenge”, God V gives predictions from several relatively abstract dimensions and perspectives:

• AI as a participant in the game

• AI as a gaming interface

• AI as the rules of the game

• AI as the game goal

We will summarize the AI ​​projects currently seen in the primary market from a more specific and direct perspective.Most of the AI+Crypto projects are based on the core of Crypto, namely “decentralization in technology (or politics) + assetization in business”.

There is nothing to say about decentralization, Web3… According to the asset-based categories, they can be roughly divided into three main tracks:

• Assetting of computing power

• Assetting of the model

• Assetting of data

• Computing power assetization

This is a relatively dense track, because in addition to various new projects, there are also Pivot for many old projects, such as Akash on Cosmos, Nosana on Solana on Pivot, and the tokens are rising rapidly after Pivot.Indirectly reflects the market’s optimism about the AI ​​track. Although RNDR focuses on decentralized rendering, it can actually serve AI, so many categories also divide RNDR into the AI ​​track

Computing power assetization can be further subdivided into two directions according to the purpose of computing power:

• One is the “decentralized computing power used for AI training” represented by Gensyn;

• One is the “decentralized computing power used for AI reasoning” represented by most Pivot and new projects;

You can see a very interesting phenomenon on this track, or you don’t like the contempt chain:

• Traditional AI → Decentralized Inference → Decentralized Training

• Traditional AI professionals are not optimistic about decentralized AI training Or reasoning

• Decentralized reasoning is not optimistic about decentralized training

The reason is mainly technically, because AI training (specifically refers to large model AI) involves massive data, and what is more exaggerated than the data demand is the bandwidth demand formed by high-speed communication of these data.In the current Transformer big model environment, training these big models requires a large number of high-end graphics cards of 4090 level/H100 professional AI graphics cards purchased from computing power matrix + NVLink and professional fiber optic switching mechanism.Say this thing can be implemented in a decentralized manner, hmm…

The demand for AI inference on computing power and communication bandwidth is much smaller than that of AI training. The possibility of decentralization is naturally much greater than that of training. This is why most computing power-related projects are inference, and basically only Gensyn is the training., a big player like Together who has raised over 100 million yuan.But similarly, from the perspective of cost-effectiveness and reliability, at least at this stage, centralized computing power is still far better than decentralization.

This is not difficult to explain why decentralized inference thinks “you can’t do it at all”, while traditional AI thinks “the training technology is unrealistic” and “inference is not reliable in business” when looking at decentralized training and reasoning.”.

Some people say that when BTC/ETH first came out, everyone also said that all distributed nodes were calculated once. This model is not reliable than cloud computing. In the end, it was also done?It depends on the future demand for correctness, unchangeable and redundant dimensions of AI training and AI reasoning. It is indeed impossible to be better than centralization by simply competing for performance, reliability, and price.

Assetting of the model

This is also a track where projects are crowded, and it is also a track that is easier to understand compared to computing power assetization, because one of the most well-known applications after ChatGPT became popular is Character.AI.You can ask questions about teaching with predecessors like Socrates and Confucius, or chat with celebrities like Musk and Ultraman Sam, and you can also talk about love with virtual idols like Hatsune Miku and General Thunder. All of this,They are all the charms of big language models.The concept of AI Agent is deeply rooted in people’s hearts through Character.AI.

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

Isn’t this AI X Crypto?!

So it is not so much about the assetization of the model, but rather the assetization of the Agent created based on the big model. After all, the big model itself cannot be put on the chain. It is more about the mapping of the Agent on the model into NFT to create a class “model” based on the mapping of the Agent on the model.The AI ​​X Crypto is visual.

There are agents in the circle who can teach you how to learn English, and there are agents who can fall in love with you. There are various types of agent searches and derivative projects such as Market Place.

The common problem in this track is that there is no technical barrier, basically the NFTization of Character.AI. Our In-House technology masters use existing open source tools and frameworks to create a speech like BMAN in one night, and the sound is also likeBMAN Agent.Second, the degree of integration with blockchain is very light, a bit like Gamefi NFT on ETH. In essence, what is stored in Metadata may be just a URL or hash. The model/Agent is all on the cloud server, and the transactions on the chain are onlyJust one ownership.

The assetization of the model/Agent will still be one of the most important tracks of AI x Crypto in the visible future. I hope that projects with relatively certain technical barriers and more closely integrated with blockchain itself can be seen in the future.Appear.

Assetting of data

Data assetization is logically the most suitable for AI+Crypto, because traditional AI training can only use visible data on the Internet, or it is more accurate – data on public domain traffic, which may account forMore 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 models, we can definitely be in various vertical fields.Have a more professional Agent/Bot.

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

Then, through AI+Crypto, under the guidance of decentralized incentives, it is a very logical approach to release data on personal and personal traffic, and assetize it, providing a better and richer “food ration” to the big model.There are indeed several teams who are deeply involved in this field.

However, the biggest difficulty in this track is that data is difficult to standardize like computing power.Decentralized computing power What model of your graphics card can be converted into directly, and it is difficult to measure the quantity, quality, purpose of private data and other dimensions. If decentralized computing power is ERC20, then decentralizedThe assetization of AI training data is a bit like ERC721, and it is also many projects of PunkAzuki, and many Traits are mixed together. The difficulty of liquidity and market is not a little bit more difficult than ERC20, so it is currently used to do AI data.Assetting projects are a bit difficult.

Another thing worth mentioning in the data track is decentralized labeling. Data assetization is used in the step of “data collection”. The collected data needs to be processed before feeding to AI. This is the data annotation.step.Currently, this step is mostly centralized man-intensive labor. Through decentralized token rewards, you can turn this Labour Work into decentralized, label it to Earn, or use it in the same way as a crowdsourcing platform., also a way of thinking.I have seen a small number of teams currently working in this field.

3. AI+Crypto missing puzzle

Let’s briefly talk about the puzzle that is missing from our perspective.

First, technical barriers.As mentioned earlier, most AI+Crypto projects have almost no barriers compared to traditional AI projects in Web2. They rely more on economic models and token incentives to spend time on the front-end experience, market and operations. Of course, this isIt is understandable that decentralization and value distribution are the strengths of Web3, but lack of core barriers will inevitably lead to the instant visual sense of X to Earn.I still look forward to more teams with core technologies in parent company OTOY like RNDR to show their strengths in Crypto.

The second is the current situation of practitioners.As far as we have observed, some teams of entrepreneurs in the AI ​​X Crypto track understand AI very well, but they don’t have a deep understanding of Web3.Some teams are very Crypto Native, but have relatively shallow achievements in the field of AI.This is very similar to the early Gamefi track. Either you understand games and think about the changes in the Web2 game chain, or you understand Web3 and think about the innovation and optimization of various gold-making models.Matr1x is the first team we encountered in Gamefi track to understand dual A on game and Crypto. This is why I wrote that Matr1x is one of the three projects I have made in 23 years of “conclusion” and we look forward to it.You can see teams who understand dual A in the fields of AI and Crypto in 24 years.

The third is business scenario.AI X Crypto is in an extremely early stage of exploration. The above-mentioned assetization types are only in a few major directions, and each direction has a track that can be carefully explored and subdivided.The combination of AI and Crypto is somewhat “black” or “rough” and does not exert the optimal competitiveness or composability of AI or Crypto, which is also in line withThe second point mentioned above is closely related.For example, our In House R&D team thought of and designed a better combination method. Unfortunately, after watching so many projects in the AI ​​track, we still haven’t seen any team entering this subdivided field, so we can only continue to wait.

What, you ask why we, a VC, think of certain scenarios first than entrepreneurs on the market?Because there are 7 great masters in our In House AI team, 5 of which are from professional AI.As for the ABCDE team’s understanding of Crypto, you know…

Finally, I want to say that although from the perspective of the primary market, AI x Crpyto is still very early and immature, this does not prevent us from optimistic that AI X Crypto will become this bull market in 24-25 years.One of the main tracks.After all, is there any better way to combine AI to liberate productivity and blockchain to liberate production relations? 🙂