What opportunities do investors see in the combination of AI and Web3?

Author: Laobai

Last time I finished writing about the BTC ecosystem, I was supposed to follow up with the article on NFT and NFTFI. However, NFTs have really cooled off recently—not just in the secondary market, but I haven't discussed any NFT or NFTFI-related projects for about two months. On the other hand, AI projects have been emerging like a geyser, so I'll continue to postpone the NFT article and bring up the trend of AI + Web3 first.

It might be too long for a Twitter thread, so I’ll post a part of it today.

1. Let's Talk About AI Itself

The AI industry was actually on the verge of cooling down. As you know, the founder of Near, Yilong, used to work in AI and was a major code contributor to TensorFlow (the most popular machine learning framework). It’s speculated that he saw no hope in AI (pre-large models) and shifted to Web3.

Then, at the end of last year, the industry welcomed ChatGPT 3.5, which revitalized the field because this time it was a qualitative change, not just the previous waves of hype and quantitative changes. A few months later, the wave of AI startups also reached our Web3. In Silicon Valley's Web2, competition became fierce, with various capital FOMO, numerous homogenized solutions starting price wars, and big companies competing with large models…

However, it’s important to note that after more than half a year of explosive growth, AI has also entered a relatively bottleneck period. For example, Google’s search interest in AI has plummeted, the user growth of ChatGPT has significantly slowed, and the randomness of AI output limits many practical scenarios… In short, we are still very far from the legendary "AGI - General Artificial Intelligence."

Currently, the Silicon Valley venture capital circle has a few judgments about the next steps for AI development:

1. There are no vertical models, only large models + vertical applications (which we will mention again when discussing Web3 + AI).

2. Data from edge devices, such as mobile phones, may become a barrier, and AI based on edge devices could also be an opportunity.

3. The length of context may trigger a qualitative change in the future (currently using vector databases as AI memory, but the context length is still insufficient).

2. Web3 + AI

AI and Web3 are actually two completely different fields. AI requires centralized computing power + massive data for training, which is very centralized, while Web3 emphasizes decentralization, so it’s not easy to combine them. However, the narrative that AI changes productivity and blockchain changes production relationships is deeply ingrained, so there will always be people seeking that intersection. In the past two months, I’ve discussed no less than 10 AI projects.

Before discussing new intersection tracks, let’s talk about the old AI + Web3 projects, which are basically platform-based, represented by FET and AGIX. How should I put it? A friend of mine who specializes in AI in China told me, "Those who used to work in AI are basically useless now, whether in Web2 or Web3; many are burdens rather than experience. The direction and future lie in large models based on Transformers like OpenAI, which have saved AI." You can ponder that.

So, general platform models are not what he sees as the promising Web3 + AI model. The more than 10 projects I’ve discussed indeed lack this aspect. Currently, the main tracks I see are as follows:

1. Bot/Agent/Assistant model assetization

2. Computing power platforms

3. Data platforms

4. Generative AI

5. DeFi trading / auditing / risk control

6. ZKML

Today, I will mainly discuss the first one, which is the assetization of Bot/Agent/Assistant, as this is the most talked about and the most homogenized track.

In simple terms, these projects mostly use OpenAI as the underlying technology, combined with other open-source or self-developed techniques, such as TTS (Text to Speech), along with specific data, to fine-tune some robots that are "better than ChatGPT" in certain fields.

For example, you can train a beautiful teacher to teach you English, and you can choose whether she has an American accent or a London accent; her personality and chatting style can also be adjusted. This way, compared to ChatGPT's mechanical and official responses, your interaction experience will be better.

Recently, there was a DAPP called HIM, a Web3 female-oriented game featuring a virtual boyfriend, which can be considered a representative of this type.

From this idea, theoretically, you could have many Bots/Agents serving you. For instance, if you want to make boiled fish, there might be a Cooking Bot fine-tuned specifically for that field to teach you, providing answers that are more professional than ChatGPT. If you want to travel, there could be a travel assistant Bot offering various travel suggestions and planning. Or if you are a project party, you could create a Discord customer service bot to help answer community questions.

In addition to creating these "GPT-based vertical application" Bots, there are also derivative projects based on this concept. For instance, Bots can be seen as "model assetization," which is somewhat akin to the "small image assetization" of NFTs. So, the popular prompts in AI—like how different prompts in MidJourney can generate different images—could also be assetized. Different prompts during Bot training can yield different effects, so prompts themselves also possess value and can be assetized.

There are also projects focused on portal indexing and searching based on such Bots. When we have thousands of Bots, how do we find the one that suits you best? At that point, we might need a portal similar to Hao123 in the Web2 world or a search engine like Google to help you "locate."

In my personal view, the assetization of Bots (models) currently has two drawbacks + two directions.

Drawback 1 - Homogenization is too severe. This is the AI + Web3 track that users find easiest to understand, somewhat resembling NFTs with a bit of utility. Therefore, the primary market is starting to show a red ocean trend, becoming competitive, but since the underlying technology is all OpenAI, there are actually no technical barriers; everyone can only compete on design and operation.

Drawback 2 - Sometimes, initiatives like putting Starbucks membership cards on-chain, while a good attempt to break out, may not be as convenient for most users as a physical or electronic membership card. The same issue exists with Web3-based Bots. If I want to learn English or chat with someone like Musk or Socrates, wouldn't it be more appealing to use Web2's http://Character.AI?

Two directions - One is the near + mid-term, where putting models on-chain might be a thought. Currently, these models somewhat resemble ETH NFTs with small images, as most metadata points to off-chain servers or IPFS rather than being purely on-chain. Models usually range from tens to hundreds of megabytes, which means they need to be stored on servers.

However, with the rapid decline in storage prices recently (2TB SSD for 500 RMB) and the advancement of storage projects like Filecoin FVM and ETHStorage, I believe that in the next two to three years, putting models of hundreds of megabytes on-chain should not be a difficult task.

You might ask, what are the benefits of putting them on-chain? Once on-chain, models can be directly called by other contracts, making them more Crypto Native, and there will definitely be more ways to play, giving a sense of a Fully Onchain Game, as all data is chain-native. Currently, I see teams exploring this area, though it is still in a very early stage.

The other direction is mid + long-term. If you think seriously about smart contracts, the most suitable application is not human-machine interaction, but "machine-machine interaction." The AI side now has the concept of AutoGPT, creating your "virtual avatar" or "virtual assistant" that can not only chat with you but also help you execute tasks based on your requests, such as booking flights, hotels, or buying domain names to set up websites…

Would you prefer an AI assistant that operates your various bank accounts and Alipay, or one that transfers funds through a blockchain address? The answer is obvious. In the future, could there be a bunch of integrated AI assistants like AutoGPT that automatically handle C2C, B2C, and even B2B payments and settlements through blockchain and smart contracts in various task scenarios? At that time, the boundary between Web2 and Web3 would become very blurred.

I'll stop here for now, as point 1 takes up too much space. In a few days, I will briefly discuss points 2, 3, 4, 5, and 6.