AI+Web3: Towers and Squares

TL;DR

1. Web3 projects with AI concepts have become targets for capital attraction in the primary and secondary markets.

2. The opportunities of Web3 in the AI industry are reflected in: using distributed incentives to coordinate potential supply in the long tail - across data, storage, and computing; while establishing open-source models and a decentralized market for AI agents.

3. AI is mainly used in the Web3 industry for on-chain finance (crypto payments, trading, data analysis) and assisting development.

4. The utility of AI+Web3 is reflected in their complementarity: Web3 is expected to counteract AI centralization, while AI is expected to help Web3 break out of its niche.

Introduction

In the past two years, the development of AI has been like pressing the accelerator. The waves stirred up by Chatgpt have not only opened a new world of generative artificial intelligence but have also created a strong current in the Web3 field.

With the support of the AI concept, financing in the crypto market has significantly improved. In just the first half of 2024, 64 Web3+AI projects completed financing, among which the AI-based operating system Zyber365 achieved the highest financing of $100 million in Series A.

The secondary market is thriving even more. According to the cryptocurrency aggregator website Coingecko, in just over a year, the total market value of the AI sector has reached $48.5 billion, with a 24-hour trading volume nearing $8.6 billion. The positive impact brought by mainstream AI technology advancements is evident, as the average price of the AI sector surged by 151% following the release of OpenAI's Sora text-to-video model. The AI effect has also influenced one of the cryptocurrency capital-raising segments, Meme: the first AI Agent concept MemeCoin—GOAT quickly gained popularity and achieved a valuation of $1.4 billion, successfully sparking the AI Meme craze.

The research and topics surrounding AI+Web3 are equally hot, from AI+Depin to AI Memecoin, and now to the current AI Agent and AI DAO, the FOMO sentiment can no longer keep up with the speed of the new narrative rotation.

AI+Web3, this combination of terms filled with hot money, opportunities, and future fantasies, is inevitably seen by some as a marriage arranged by capital. We seem to find it hard to distinguish whether beneath this gorgeous robe lies the playground of speculators or the eve of an explosive dawn?

To answer this question, a key consideration for both parties is whether it will get better with the other involved. Can one benefit from the other's model? In this article, we also try to examine this pattern from the shoulders of our predecessors: how Web3 can play a role in various aspects of the AI technology stack, and what new vitality AI can bring to Web3?

Part.1 What opportunities does Web3 have under the AI stack?

Before diving into this topic, we need to understand the technology stack of AI large models:

In simpler terms, the entire process can be expressed as follows: a "large model" is like the human brain. In the early stages, this brain belongs to a newborn baby who needs to observe and absorb a vast amount of external information to understand the world. This is the "data collection" phase. Since computers do not possess multiple human senses such as vision and hearing, large-scale unlabelled external information needs to be transformed into a format that computers can understand and use through "preprocessing" before training.

After inputting data, the AI constructs a model with understanding and predictive capabilities through "training", which can be seen as the process of an infant gradually understanding and learning about the outside world. The parameters of the model are akin to the language abilities that the infant continuously adjusts during the learning process. When the learning content begins to be categorized, or when feedback is received through communication with others and corrections are made, it enters the "fine-tuning" stage of the large model.

As children grow up and learn to speak, they can understand meanings and express their feelings and thoughts in new conversations. This stage is similar to the "reasoning" of AI large models, where the model is able to predict and analyze new language and text inputs. Infants express feelings, describe objects, and solve various problems through their language abilities, which is also akin to how AI large models apply reasoning in various specific tasks after completing training and being put into use, such as image classification and speech recognition.

The AI Agent is moving closer to the next form of large models - capable of independently executing tasks and pursuing complex goals. It not only possesses thinking abilities but can also remember, plan, and interact with the world using tools.

Currently, in response to the pain points of AI across various stacks, Web3 has initially formed a multi-layered, interconnected ecosystem that covers all stages of the AI model process.

1. Basic Layer: Computing Power and Data's Airbnb

Hashrate

Currently, one of the highest costs of AI is the computing power and energy required for training and inference models.

One example is that a large tech company's large language model requires 16,000 high-performance GPUs produced by a well-known chip manufacturer to complete training in 30 days. The unit price of the latter's 80GB version ranges from $30,000 to $40,000, which requires an investment of $400 to $700 million in computing hardware (GPUs + network chips), while the monthly training consumes 1.6 billion kilowatt-hours, resulting in energy expenses of nearly $20 million per month.

The release of AI computing power is also the earliest intersection of Web3 and AI — DePin (Decentralized Physical Infrastructure Network). Currently, the DePin Ninja data website has listed over 1,400 projects, including representative projects in GPU computing power sharing such as io.net, Aethir, Akash, Render Network, and so on.

The main logic is that the platform allows individuals or entities with idle GPU resources to contribute their computing power in a decentralized manner without permission. By creating an online marketplace similar to a sharing economy platform for buyers and sellers, the utilization rate of underutilized GPU resources is increased, allowing end users to obtain more cost-effective and efficient computing resources. At the same time, the staking mechanism ensures that resource providers face corresponding penalties in case of violations of quality control mechanisms or network disruptions.

Its characteristics are:

• Pooling idle GPU resources: The suppliers mainly include surplus computing power resources from third-party independent small and medium-sized data centers, cryptocurrency mining farms, etc., with PoS consensus mechanism mining hardware, such as FileCoin and ETH miners. Currently, there are also projects aimed at starting with lower entry barriers, such as exolab, which utilizes local devices like MacBook, iPhone, and iPad to establish a computing network for running large model inference.

• Facing the long-tail market of AI computing power:

a. "In terms of technology, the decentralized computing power market is more suitable for inference steps. Training relies more on the data processing capabilities brought by ultra-large cluster scale GPUs, while inference has relatively lower requirements for GPU computing performance, such as Aethir focusing on low-latency rendering work and AI inference applications.

b. "From the demand side perspective," small and medium computing power demanders will not train their own large models separately, but will only choose to optimize and fine-tune around a few leading large models, and these scenarios are naturally suitable for distributed idle computing power resources.

• Decentralized ownership: The technological significance of blockchain lies in the fact that resource owners always retain control over their resources, allowing for flexible adjustments based on demand while also generating profits.

Data

Data is the foundation of AI. Without data, computation is as useless as floating weeds, and the relationship between data and models is akin to the saying "Garbage in, Garbage out." The quantity of data and the quality of input determine the final output quality of the model. For the training of current AI models, data determines the model's language capabilities, understanding abilities, even values, and human-like performance. Currently, the data demand dilemma for AI mainly focuses on the following four aspects:

• Data hunger: AI model training relies on large amounts of data input. Public information shows that a well-known AI company has trained its large language model with a parameter count reaching the trillion level.

• Data Quality: With the integration of AI and various industries, the timeliness of data, diversity of data, professionalism of vertical data, and the incorporation of emerging data sources such as social media sentiment have also raised new requirements for its quality.

• Privacy and compliance issues: Currently, various countries and enterprises are gradually recognizing the importance of high-quality datasets and are imposing restrictions on dataset scraping.

• High data processing costs: large data volume and complex processing. Public information shows that over 30% of AI companies' R&D costs are used for basic data collection and processing.

Currently, Web3 solutions are reflected in the following four aspects:

1. Data Collection: The availability of real-world data that can be scraped for free is rapidly depleting, and the spending by AI companies on data is increasing year by year. However, at the same time, this expenditure has not been reflected back to the true contributors of the data; platforms are fully enjoying the value creation brought about by the data. For example, a certain social platform generated a total revenue of $203 million through data licensing agreements with AI companies.

The vision of Web3 is to allow users who genuinely contribute to also participate in the value creation brought by data, and to obtain more private and valuable data from users in a cost-effective manner through distributed networks and incentive mechanisms.

• Grass is a decentralized data layer and network that allows users to run Grass nodes, contribute idle bandwidth and relay traffic to capture real-time data from the entire internet, and earn token rewards.

• Vana introduces a unique Data Liquidity Pool (DLP) concept, allowing users to upload their private data (such as shopping records, browsing habits, social media activities, etc.) to specific DLPs and flexibly choose whether to authorize specific third parties to use this data.

• In PublicAI, users can use #AI或#Web3 as a classification tag on a certain social platform and @PublicAI to achieve data collection.

2. Data Preprocessing: In the AI data processing process, the collected data is often noisy and contains errors, so it must be cleaned and converted into a usable format before training the model, involving repetitive tasks such as normalization, filtering, and handling missing values. This stage is one of the few manual steps in the AI industry, giving rise to the profession of data annotators. As the model's requirements for data quality increase, the threshold for data annotators also rises, and this task is inherently suitable for Web3's decentralized incentive mechanisms.

• Currently, Grass and OpenLayer are both considering incorporating data labeling as a key component.

• Synesis introduced the concept of "Train2earn," emphasizing data quality, where users can earn rewards by providing labeled data, annotations, or other forms of input.

• The data annotation project Sapien gamifies the labeling tasks and allows users to stake points to earn more points.

3. Data Privacy and Security: It is important to clarify that data privacy and security are two different concepts. Data privacy involves the handling of sensitive data, while data security protects data information from unauthorized access, destruction, and theft. Therefore, the advantages of Web3 privacy technologies and their potential application scenarios are reflected in two aspects: (1) Training on sensitive data; (2) Data collaboration: Multiple data owners can jointly participate in AI training without having to share their original data.

The currently common privacy technologies in Web3 include:

• Trusted Execution Environment ( TEE ), such as Super Protocol.

• Fully Homomorphic Encryption (FHE), such as BasedAI, Fhenix.io or Inco Network.

• Zero-knowledge technology (zk), such as Reclaim Protocol using zkTLS technology, generates zero-knowledge proofs of HTTPS traffic, allowing users to securely import activity, reputation, and identity data from external websites without exposing sensitive information.

However, the field is still in its early stages, and most projects are still exploring. One current dilemma is that the computational costs are too high, some examples are:

• The zkML framework EZKL takes about 80 minutes to generate a proof for a 1M-nanoGPT model.

• According to data from Modulus Labs, the overhead of zkML is more than 1000 times higher than pure computation.

4. Data Storage: Once the data is available, a place is needed to store the data on-chain, as well as the LLM generated using that data. With data availability (DA) as the core issue, the throughput was 0.08MB before the Ethereum Danksharding upgrade. Meanwhile, the training and real-time inference of AI models typically require a data throughput of 50 to 100GB per second. This magnitude of difference leaves existing on-chain solutions struggling when facing "resource-intensive AI applications."

• 0g.AI is a representative of this category.