Opportunities in AI and Web3: Investors’ Perspectives on Prospects and Opportunities

A combination of AI with Web3 was not recommended because AI is relatively centralized whereas Web3 is focused on decentralization.

While the NFT secondary market is declining, the primary market doesn’t seem to have talked about NFT or NFTFi-related projects in the past two months, while AI projects are really starting to take off, so the NFT article continues to drag. When I finished writing about the BTC ecology last time, I should have added the article about NFT and NFTFI. However, NFT is really not generally cool right now. The trend of combining AI with Web3 is first mentioned.

AI                            

AI industry was about to decline. Illia, the Founder of Near Protocol, is actually skilled in AI, the primary code contributor to TensorFlow, the most well-known machine learning framework. People speculate that he thought there is no hope in  AI (machine learning before the big model) and came to do Web3.

However, the industry welcomed ChatGpt3.5 around the end of last year, and the AI business began to revive. Because, unlike prior rounds of hype and quantitative change, this time may truly be considered a qualitative change. After a few months, the AI entrepreneurial tsunami has also reached Web3. The internal competition of Silicon Valley Web2 is fierce, a wide range of capital are Fomo, lots of homogenization programs that have begun to wage a price war, large factories and large models are PK…

However, it should be noted that AI has also entered a relative bottleneck period after more than half a year of outbreak, such as Google’s search popularity with AI fell off a cliff, Chatgpt user growth slowed down significantly, AI Output with certain randomness limited many landing scenes… All in all, we are very, very far from the fabled “AGI-general artificial intelligence.”

At present, the Silicon Valley Venture Capitals  take such judgments on the next development of AI:

1. There are no vertical models, only large model vertical applications (we’ll come back to that later when we talk about Web3+AI);
2. Data on edge devices such as mobile phones may be a barrier, and AI-based on-edge devices may also be an opportunity;
3. The length of Context may cause qualitative changes in the future (vector database is now used as AI memory, but the context length is still not enough).

Web3+AI

In reality, AI and Web3 are two entirely distinct technologies. AI requires concentrated processing power and enormous amounts of data to perform training, which is quite centralized. Although they don’t work well together since Web3 is decentralized, the notion that AI changes productivity and blockchain changes the production relationship is too pervasive. There will always be individuals looking for that intersection, and over the past two months, I have discussed no less than 10 AI initiatives.

Before talking about the new combination track, let’s start by discussing the old AI+Web3 projects, which are basically platform types, represented by FET and AGIX. My domestic professional AI friends told me that All this AI stuff that used to be done is basically useless now, whether Web2 or Web3, many are baggage rather than experience.” The direction and future is in big models like OpenAI based on the transformer, big models saved AI.

Therefore, the general platform type is not the model of Web3+AI that he is optimistic about, and the more than 10 projects I talked about do not have this aspect, and the basic things I see are the following sectors:

1. Bot/Agent/Assistant Capitalization
2. Computing platform
3. Data Platform
4. Generative AI
5. DeFi Trading/Auditing/Risk Control
6. ZKML
7. Bot/Agent/Assitant capitalization

The most discussed and homogenized sector is this one. Simply expressed, these projects typically use OpenAI as the foundation, along with additional open source or self-developed technical tools like TTS (Text to Speech), and with specific data, FineTune creates some bots that are “better than ChatGPT in a certain field.”

For instance, you may hire a gorgeous instructor to teach you English.She has a choice between a Cockney accent and an American one. Your interaction experience will be better than it would be with ChatGPT, which is more formal and mechanical, thanks to the ability to modify her personality and chat style. there is a virtual boyfriend DAPP, a Web3 women game, called HIM, which can be regarded as a representative of this type.

From this concept, you can theoretically have many bots/agents to serve you. For instance, if you want to learn how to boil fish, there may be a Fine Tune Cooking Bot specifically for this field to teach you, and the response is more qualified than ChatGPT. If you want to travel, there are also travel bots that can give you travel advice and planning. Or, if you’re a project partner, get a Discord customer service robot to help you respond to community questions.

There are derivative projects based on this kind of “GPT-based vertical application type” Bot in addition to doing this type of “GPT-based vertical application type” Bot, as model capitalization” and NFT as picture capitalization,” which means that now the well-known Prompt in AI may also be capitalized. Promopt itself has value and can be capitalized, for instance, different MidJourney prompts can produce distinct visuals and varied prompts will have different results when training bots.

On such bots, there are additional initiatives like portal indexing and searching. How can we choose the best Bot for you when we have thousands of them? Then, to assist you in “locating,” you might require a Web2 portal like Hao123 or a search engine like Google.

I believe that at this time, Bot (model) capitalization has two drawbacks and two directions:

Drawback 1. Because this is the most user-friendly AI+web3 track, homogenization is a major drawback. There are elements of NFT with hints of utility characteristics. As a result, the primary market has started to exhibit the Red Sea trend and compete, but OpenAI is at the bottom, so we genuinely have no technical obstacles to overcome; we can only compete on the basis of design and operation.

Drawback 2. A physical or electronic membership card may be more convenient for most users, despite the fact that Starbucks membership card NFT on-Chain makes a commendable effort to enter the world of web3. This is also a problem with Web3-based bots as well. If I want to converse with Musk, Socrates, or the robot to learn English, Would it not be beneficial for me to utilize Web2 http://Character.AI right away?

Direction 1.  Model on-chain might be a good concept in the short- to medium term. These models currently have a limited understanding of the ETH NFT, with MetaData primarily pointing to IPFS or off-chain servers rather than the blockchain itself. Especially on the server, models are typically tens to hundreds of megabytes in size.

However, I think that it shouldn’t be challenging to chain models in the 100 megabits scale in the next two or three years given the recent significant reduction in storage prices (2TB SSD 500RMB) and the development of storage projects like Filecoin FVM and ETHStorage.

What are the advantages of on-chain, you might wonder? The on-chain model may be directly used by other contracts, which is more Crypto Native, and the game patterns are unquestionably greater. There’s a bit of a visual sense of a Fully Onchain Game because all the data is native to the chain. there are many teams currently exploring this, however, it is still very early in the process.

Direction2. If you give smart contracts some serious thought over the medium- long term, you may find that “machine-machine interaction” rather than human-computer interaction

is more appropriate. The AI now has the idea of AutoGPT, which allows you to get a “virtual avatar” or “virtual assistant” who can assist you with tasks such as booking tickets, hotels, purchasing a domain name to create a website, and other items as needed.

Do you prefer the convenience of Alipay and all kinds of bank cards or the convenience of the complete blockchain address transfer when it comes to having the AI assistant manage your numerous bank accounts? The solution is clear. Will there, therefore, be a plethora of AI helpers like AutoGPT in the future that carry out C2C, B2C, or even B2B payment and settlement automatically through blockchain and smart contracts in a variety of task scenarios? Then the line separating Web2 and Web3 got pretty hazy at that moment.

2. Computing Platform

The computing power platform project is less capitalized and competitive compared to the Bot model, but it is relatively easier to comprehend. AI requires significant computing power, and BTC and ETH have proven in the last decade that such a method exists, which can organize and coordinate massive computing power to cooperate and compete in a decentralized environment of economic incentives and games. Now, this approach can be applied to AI.

Together and Gensyn are without a doubt the two most well-known projects in the sector,  one received $10 million in seed funding and the other $43 million in the A round. These two are trying to raise a lot of money since they require money and processing capacity to build their own models first, after which they will use those platforms to build other AI projects.

The amount of financing for the reasoning computing power platform will be relatively small because it is essentially the aggregation of idle GPU and other computing power and then provided to the AI project in need of reasoning. RNDR does render power aggregation, and these platforms do inference power aggregation. But the technical threshold is relatively vague at present, and I even wonder if one day RNDR or Web3 cloud computing platform will extend to the reasoning computing platform.

The direction of the computing power platform is more realistic and a better predictor than the model capital; basically, there will be a need for it and one or two leading projects to see who can do it. The only uncertainty is whether training and reasoning have separate leading projects, or whether the leading projects will do both.

3. Data Platform

This is not difficult to understand, because the underlying AI is three major things: algorithms (models), computing power and data.

Since the algorithm and computing power have a “decentralized version”, the data will certainly not be absent, which is also the most optimistic direction of Dr. Lu Qi, the founder of the Qiji creation platform, when talking about AI and Web3.

Web3 has always emphasized data privacy and sovereignty, and there are technologies such as ZK to ensure data reliability and integrity, so the AI trained based on Web3 on-chain data must be different from that trained on Web2 off-chain data. So this direction as a whole makes sense. Ocean should be considered belong to this sector, and there is also projects such as specialized AI data markets based on Ocean in the primary market

4. Generative AI

Simply said, it involves using AI painting or other comparable production to support several other scenes, including the construction of NFT, in-game maps, NPC backgrounds, and so forth. This method of implementing NFT is challenging because the scarcity of AI generation is insufficient. But  Gamefi may be possible, there are teams attempting to implement Gamefi in the primary market.

However, a few days ago, Unity (together with Unreal Engine, which has long dominated the game engine market) also released Sentis and Muse, two of its own AI-generating tools. These tools are now in the limited beta stage, but they are anticipated to be formally released next year. Web3 game AIGC projects, maybe dimension decrease hit by Unity…

5. DeFi Trading/Auditing/Risk Control

Projects have been attempted in these categories, but homogenization is not very evident.

• Trading – This one is Tricky because if a trading strategy works well, as more people use it, it may become less useful and you have to switch to a new strategy. Then we are curious about the future win rate of AI trading robots and what position they will be in among ordinary traders.
• Audit – It should assist in promptly addressing existing common vulnerabilities, but not for logical or brand-new weaknesses, and this should only occur in the AGI age.
• Yield – Yield is simple to understand. Just picture a YFI with AI intelligence and invest your funds there. AI Staking will choose platforms to stake, make LP pools, and mine based on your risk tolerance.
• Risk control – it feels strange to do a project alone, and it makes sense to service various lending or Defi platforms in the form of plug-ins.

6. ZKML

This is the sector that is becoming increasingly popular Because it combines two of the most cutting-edge technologies, ZK and ML (machine learning, a specialized area of AI).

In theory, the combination of ZK can give ML privacy, integrity, and accuracy, but in practice, many project parties struggle to come up with specific use scenarios and instead focus on building infrastructure.

The only thing that’s really needed right now is machine learning in parts of the medical field, the need for the privacy of patient data, and narratives like on-chain game integrity or anti-cheating always feel far-fetched.

Modulus Labs, EZKL, Giza, etc., are some of the hottest projects in this sector in the primary market.

The technical barrier of this sector is considerably higher than other tracks and the homogenization is generally not visible. There aren’t many people in the world that understand ZK, and there are even fewer talents who comprehend ZK and ML. ZKML focuses more on reasoning than training.

So much for the trends of AI + Web 3. If you have any ideas for projects that would be a good fit but fall outside of my list of the top six sectors, please message me.

Original Link: https://twitter.com/Wuhuoqiu/status/1676817186831360001?s=20

https://twitter.com/Wuhuoqiu/status/1679370185369874433?s=20