Author: Coinspire
►TL;DR
1. Web3 projects with AI concepts have become gold-absorbing targets 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; at the same time, establishing an open source model and a decentralized market for AI Agents.
3. AI is mainly used in the Web3 industry for on-chain finance (encrypted payments, transactions, data analysis) and auxiliary development.
4. The effectiveness of AI+Web3 is reflected in their complementarity: Web3 is expected to fight against AI centralization, and AI is expected to help Web3 break the circle.
introduction
In the past two years, the development of AI seems to have been accelerated. The butterfly wings fanned by Chatgpt not only opened up a new world of generative artificial intelligence, but also set off an ocean current in Web3 on the other side.
With the support of AI concept, the financing of the crypto market, which has slowed down, has been significantly boosted. According to media statistics, in the first half of 2024 alone, a total of 64 Web3+AI projects completed financing, and the artificial intelligence-based operating system Zyber365 achieved the highest financing amount of US$100 million in the A round.
The secondary market is even more prosperous. Data from the crypto aggregation website Coingecko shows that in just over a year, the total market value of the AI sector has reached US$48.5 billion, with a 24-hour trading volume of nearly US$8.6 billion. The benefits brought by the progress of mainstream AI technology are obvious. After the release of OpenAI's Sora text-to-video model, the average price of the AI sector rose by 151%. The AI effect also radiates to Meme, one of the money-making sectors of cryptocurrency: MemeCoin, the first AI Agent concept, quickly became popular and gained a valuation of US$1.4 billion, successfully setting off an AI Meme craze.
Research and topics about AI+Web3 are equally hot. From AI+Depin to AI Memecoin to the current AI Agent and AI DAO, FOMO sentiment can no longer keep up with the speed of new narrative rotation.
AI+Web3, this term combination full of hot money, hot spots and future fantasies, is inevitably seen as an arranged marriage brokered by capital. It seems difficult for us to distinguish under this gorgeous robe, whether it is the home ground of speculators or the eve of the dawn of an explosion?
To answer this question, a key question for both parties is, will it be better with the other party? Can it benefit from the other party's model? In this article, we also try to stand on the shoulders of our predecessors to examine this situation: how can Web3 play a role in each link of the AI technology stack, and what new vitality can AI bring to Web3?
Part.1 What opportunities does Web3 have under the AI stack?
Before we start this topic, we need to understand the technology stack of the AI big model:
Image source: Delphi Digital
To describe the whole process in more popular language: the "big model" is like the human brain. In the early stages, this brain belongs to a newborn baby, which needs to observe and absorb massive amounts of external information around it to understand the world. This is the "collection" stage of data. Since computers do not have multiple senses such as human vision and hearing, before training, large-scale unlabeled information from the outside world needs to be converted into an information format that the computer can understand and use through "preprocessing."
After the data is input, AI builds a model with understanding and prediction capabilities through "training", which can be seen as the process of infants gradually understanding and learning the outside world. The parameters of the model are like the language ability that infants constantly adjust during the learning process. When the learning content begins to be divided into subjects, or when communicating with people to get feedback and make corrections, it enters the "fine-tuning" stage of the big model.
As children grow up and learn to speak, they can understand the meaning and express their feelings and thoughts in new conversations. This stage is similar to the "reasoning" of large AI models, which can predict and analyze new language and text inputs. Babies use language to express their feelings, describe objects, and solve various problems. This is also similar to the application of large AI models to various specific tasks in the reasoning stage after they are trained and put into use, such as image classification and speech recognition.
AI Agent is closer to the next form of the big model - it can perform tasks independently and pursue complex goals. It not only has the ability to think, but also can remember, plan, and use tools to interact with the world.
Currently, in response to the pain points of AI in various stacks, Web3 has initially formed a multi-level, interconnected ecosystem that covers all stages of the AI model process.
1. Basic layer: Airbnb of computing power and data
▎Computing power
Currently, one of the highest costs of AI is the computing power and energy required to train and infer models.
For example, Meta's LLAMA3 requires 16,000 H100 GPUs produced by NVIDIA (a top-level graphics processing unit designed for artificial intelligence and high-performance computing workloads) for 30 days to complete training. The unit price of the latter 80GB version is between US$30,000 and US$40,000, which requires US$400 million to US$700 million in computing hardware investment (GPU + network chip). At the same time, monthly training consumes 1.6 billion kWh, and energy expenditure is nearly US$20 million per month.
The decompression of AI computing power is also the earliest area where Web3 intersects with AI - DePin (decentralized physical infrastructure network). Currently, the DePin Ninja data website has displayed more than 1,400 projects, among which representative projects of GPU computing power sharing include io.net, Aethir, Akash, Render Network, etc.
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, and through an online marketplace for buyers and sellers similar to Uber or Airbnb, it increases the utilization rate of under-utilized GPU resources, so that end users can obtain more efficient computing resources at a lower cost. At the same time, the pledge mechanism also ensures that if there is a violation of the quality control mechanism or network interruption, the resource provider will be punished accordingly.
Its characteristics are:
Aggregate idle GPU resources: The suppliers are mainly third-party independent small and medium-sized data centers, crypto mining farms and other operators' excess computing resources, and mining hardware with PoS consensus mechanism, such as FileCoin and ETH mining machines. Currently, there are also projects dedicated to launching devices with lower thresholds, such as exolab, which uses local devices such as MacBook, iPhone, iPad, etc. to establish a computing network for running large model inference.
Facing the long-tail market of AI computing power:
a. "From a technical perspective" the decentralized computing power market is more suitable for the inference step. Training relies more on the data processing capabilities brought by super-large cluster-scale GPUs, while inference has relatively low GPU computing performance. For example, Aethir focuses on low-latency rendering work and AI inference applications.
b. "On the demand side" Small and medium-sized computing power demanders will not train their own large models separately, but will only choose to optimize and fine-tune a few top large models. These scenarios are naturally suitable for distributed idle computing power resources.
Decentralized ownership: The technical significance of blockchain is that resource owners always retain control over their resources, flexibly adjust them according to demand, and obtain benefits at the same time.
Data
Data is the foundation of AI. Without data, computing is useless, and the relationship between data and models is like the proverb "Garbage in, Garbage out". The quantity and quality of data input determine the output quality of the final model. For the training of current AI models, data determines the model's language ability, understanding ability, and even values and human performance. At present, AI's data demand dilemma mainly focuses on the following four aspects:
Data Hungry: AI model training relies on a large amount of data input. Public data shows that the number of parameters used by OpenAI to train GPT-4 has reached trillions.
Data quality: With the integration of AI into various industries, new requirements have been put forward for data timeliness, data diversity, professionalism of vertical data, and the intake of emerging data sources such as social media sentiment.
Privacy and compliance issues: Currently, various countries and companies are gradually realizing the importance of high-quality data sets and are restricting data set crawling.
High data processing costs: large amounts of data and complex processing. Public data shows that more than 30% of AI companies’ R&D costs are spent on basic data collection and processing.
At present, web3's solutions are reflected in the following four aspects:
1. Data collection: The real-world data that can be captured for free is rapidly running out, and AI companies are spending more and more money on data each year. But at the same time, this expenditure has not been fed back to the real contributors of the data, and the platform has fully enjoyed the value creation brought by the data. For example, Reddit has achieved a total revenue of US$203 million through data licensing agreements signed with AI companies.
The vision of Web3 is to allow users who truly contribute to participate in the value creation brought by data, and to obtain users' more private and valuable data in a low-cost manner through distributed networks and incentive mechanisms.
For example, Grass is a decentralized data layer and network. Users can run Grass nodes, contribute idle bandwidth and relay traffic to capture real-time data from the entire Internet, and obtain token rewards.
Vana introduces a unique concept of Data Liquidity Pool (DLP), where users can upload their private data (such as shopping records, browsing habits, social media activities, etc.) to a specific DLP and flexibly choose whether to authorize the use of this data to specific third parties;
In PublicAI, users can use #AI or #Web3 as classification tags on X and @PublicAI to collect data.
2. Data preprocessing: In the data processing process of AI, since the collected data is usually noisy and contains errors, it must be cleaned and converted into a usable format before training the model, involving repetitive tasks of standardization, filtering, and processing missing values. This stage is one of the few manual links in the AI industry, and has spawned the industry of data labelers. As the model's requirements for data quality increase, the threshold for data labelers also increases, and this task is naturally suitable for the decentralized incentive mechanism of Web3.
Currently, both Grass and OpenLayer are considering adding data labeling, a key step.
Synesis proposed the concept of "Train2earn", which emphasizes data quality. Users can get rewards by providing labeled data, annotations or other forms of input.
The data annotation project Sapien gamifies the labeling task and allows users to stake points to earn more points.
3. Data privacy and security: It needs to be clarified that data privacy and security are two different concepts. Data privacy involves the processing of sensitive data, while data security protects data information from unauthorized access, destruction, and theft. Therefore, the advantages and potential application scenarios of Web3 privacy technology are reflected in two aspects: (1) training of sensitive data; (2) data collaboration: multiple data owners can participate in AI training together without sharing their original data.
The more common privacy technologies in Web3 currently include:
Trusted Execution Environment (TEE), such as Super Protocol;
Fully homomorphic encryption (FHE), such as BasedAI, Fhenix.io, or Inco Network;
Zero-knowledge (zk) technologies, such as Reclaim Protocol, use zkTLS technology to generate zero-knowledge proofs for 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 under exploration. One of the current difficulties is that the computational cost is too high. Some examples are:
The zkML framework EZKL takes about 80 minutes to generate a proof for a 1M-nanoGPT model.
According to Modulus Labs, zkML’s overhead is more than 1,000 times higher than pure computation.
4. Data storage: After you have the data, you also need a place to store it on the chain, as well as the LLMs generated using that data. With data availability (DA) as the core issue, before the Ethereum Danksharding upgrade, its throughput was 0.08MB. At the same time, AI model training and real-time reasoning typically require 50 to 100GB of data throughput per second. This order of magnitude gap makes existing on-chain solutions unable to cope with "resource-intensive AI applications."
0g.AI is a representative project in this category. It is a centralized storage solution designed for AI high-performance requirements. Its key features include high performance and scalability. Through advanced sharding and erasure coding technologies, it supports fast upload and download of large-scale data sets, with a data transfer speed of nearly 5GB per second.
2. Middleware: Model Training and Inference
▎Open Source Model Decentralized Market
The debate over whether AI models should be closed source or open source has never disappeared. The collective innovation brought by open source is an advantage that closed source models cannot match. However, under the premise of no profit model, how can open source models improve the driving force of developers? This is a direction worth thinking about. Baidu founder Robin Li once asserted in April this year that "open source models will fall further and further behind."
In response to this, Web3 proposes the possibility of a decentralized open source model market, which is to tokenize the model itself, reserve a certain proportion of tokens for the team, and flow part of the future income of the model to token holders.
For example, the Bittensor protocol establishes an open source P2P market consisting of dozens of "subnets" where resource providers (computing, data collection/storage, machine learning talent) compete with each other to meet the goals of specific subnet owners. Subnets can interact and learn from each other to achieve more powerful intelligence. Rewards are allocated by community voting and further distributed among subnets based on competition performance.
ORA introduces the concept of Initial Model Offering (IMO), tokenizing AI models so that they can be bought, sold, and developed through a decentralized network.
Sentient, a decentralized AGI platform, incentivizes human contributors to collaborate, build, replicate, and scale AI models, and rewards contributors.
Spectral Nova, focusing on the creation and application of AI and ML models.
Verifiable Reasoning
The standard Web3 solution to the “black box” problem in AI’s reasoning process is to have multiple validators repeat the same operations and compare the results, but due to the current shortage of high-end “Nvidia chips”, the obvious challenge facing this approach is the high cost of AI reasoning.
A more promising solution is to perform ZK proofs (zero-knowledge proofs, a cryptographic protocol in which one party, a prover, can prove to another party, a verifier, that a given statement is true without revealing any additional information other than that the statement is true) on the off-chain AI reasoning computation, and perform permissionless verification of AI model computations on the chain. This requires cryptographically proving on the chain that the off-chain computation has been completed correctly (for example, the data set has not been tampered with) while ensuring that all data is kept confidential.
Key benefits include:
Scalability: Zero-knowledge proofs can quickly confirm large amounts of off-chain computations. Even if the number of transactions increases, a single zero-knowledge proof can verify all transactions.
Privacy protection: Data and AI model details remain private, while all parties can verify that the data and models have not been compromised.
Trustless: No reliance on centralized parties to confirm computations.
Web2 Integration: Web2 is off-chain integrated by definition, which means that verifiable reasoning can help bring its datasets and AI computations on-chain. This helps increase the adoption of Web3.
The current verifiable technologies for verifiable reasoning in Web3 are as follows:
zkML: Combines zero-knowledge proof with machine learning to ensure the privacy and confidentiality of data and models, allowing verifiable computations without revealing certain underlying properties. For example, Modulus Labs has released a ZK prover built for AI based on ZKML to effectively check whether AI providers manipulate algorithms on the chain to execute correctly. However, current customers are basically on-chain DApps.
opML: Utilizes the optimistic rollup principle to improve the scalability and efficiency of ML computations by verifying when disputes occur. In this model, only a small portion of the results generated by the “verifier” needs to be verified, but the economic cost of cutting is set high enough to increase the cost of cheating for the verifier and save redundant computations.
TeeML: Securely execute ML computations using a trusted execution environment, protecting data and models from tampering and unauthorized access.
3. Application layer: AI Agent
The current development of AI has already shown a transition of development focus from model capabilities to AI Agents. OpenAI, AI big model unicorn Anthropic, Microsoft and other technology companies have turned to the development of AI Agents, trying to break the current LLM technology platform period.
OpenAI defines AI Agent as a system that uses LLM as its brain, has the ability to autonomously understand, perceive, plan, remember, and use tools, and can automatically perform complex tasks. When AI changes from a tool to a subject that can use tools, it becomes an AI Agent. This is why AI Agent can become the most ideal intelligent assistant for humans.
And what can Web3 bring to Agent?
▎Decentralization
The decentralized nature of Web3 can make the Agent system more decentralized and autonomous. By establishing incentive and punishment mechanisms for pledgers and delegators through mechanisms such as PoS and DPoS, the democratization of the Agent system can be promoted. GaiaNet, Theoriq, and HajimeAI have all tried this.
Cold Start
The development and iteration of AI Agents often require a lot of financial support, and Web3 can help potential AI Agent projects obtain early financing and cold start.
Virtual Protocol launches the AI Agent creation and token issuance platform fun.virtuals, where any user can deploy AI Agent with one click and achieve 100% fair issuance of AI Agent tokens.
Spectral proposed a product concept that supports the issuance of on-chain AI Agent assets: by issuing tokens through IAO (Initial Agent Offering), AI Agent can obtain funds directly from investors and become a member of DAO governance, providing investors with the opportunity to participate in project development and share future profits.
Part.2 How does AI empower Web3?
The impact of AI on Web3 projects is obvious. It benefits blockchain technology by optimizing on-chain operations (such as smart contract execution, liquidity optimization, and AI-driven governance decisions). At the same time, it can also provide better data-driven insights, improve on-chain security, and lay the foundation for new Web3-based applications.
1. AI and on-chain finance
▎AI and Crypto Economy
On August 31, Coinbase CEO Brian Armstrong announced the first AI-to-AI crypto transaction on the Base network, and said that AI Agents can now use USD on Base to trade with humans, merchants or other AIs. These transactions are instant, global, and free.
In addition to payment, Virtuals Protocol's Luna also demonstrated for the first time how AI Agent can autonomously perform on-chain transactions in the following ways, which has attracted attention. AI Agent, as an intelligent entity that can perceive the environment, make decisions and execute actions, is regarded as the future of on-chain finance. At present, the potential scenarios of AI Agent are reflected in the following points:
1. Information collection and prediction: Help investors collect information such as exchange announcements, project public information, panic sentiment, public opinion risks, etc., analyze and evaluate asset fundamentals and market conditions in real time, and predict trends and risks.
2. Asset management: provide users with suitable investment targets, optimize asset portfolios, and automatically execute transactions.
3. Financial experience: Help investors choose the fastest on-chain transaction method, automate manual operations such as cross-chain and gas fee adjustment, and reduce the threshold and cost of on-chain financial activities.
Imagine a scenario where you give the following instructions to the AI Agent: "I have 1,000 USDT. Please help me find the combination with the highest return, with a lock-up period of no more than one week." The AI Agent will provide you with the following suggestions: "The recommended initial allocation is 50% in A, 20% in B, 20% in X, and 10% in Y. I will monitor the interest rate and observe changes in its risk level and rebalance when necessary." In addition, looking for potential airdrop projects, as well as Memecoin projects that show signs of a popular community, are all things that the AI Agent may be able to achieve later.
Image source: Biconomy
Currently, AI Agent wallet Bitte and AI interaction protocol Wayfinder are both making such attempts. They are trying to access OpenAI's model API, allowing users to command Agents to complete various on-chain operations in a chat window interface similar to ChatGPT. For example, WayFinder's first prototype released in April this year demonstrated the four basic operations of swap, send, bridge and stake on the three public chain mainnets of Base, Polygon and Ethereum.
Currently, the decentralized agent platform Morpheus also supports the development of such agents. For example, Biconomy has also demonstrated an operation in which an AI agent can swap ETH into USDC without authorizing full wallet permissions.
▎AI and on-chain transaction security
In the Web3 world, on-chain transaction security is crucial. AI technology can be used to enhance the security and privacy protection of on-chain transactions. Potential scenarios include:
Transaction monitoring: Real-time data technology monitors abnormal transaction activities and provides real-time alert infrastructure for users and platforms.
Risk analysis: helps the platform analyze customer transaction behavior data and assess its risk level.
For example, the Web3 security platform SeQure uses AI to detect and prevent malicious attacks, fraud, and data leaks, and provides real-time monitoring and alert mechanisms to ensure the security and stability of on-chain transactions. Similar security tools include AI-powered Sentinel.
2. AI and on-chain infrastructure
▎AI and on-chain data
AI technology plays an important role in on-chain data collection and analysis, such as:
Web3 Analytics: is an AI-based analytics platform that uses machine learning and data mining algorithms to collect, process and analyze on-chain data.
MinMax AI: It provides AI-based on-chain data analysis tools to help users discover potential market opportunities and trends.
Kaito: A Web3 search platform based on LLM's search engine.
Followin: It integrates ChatGPT to collect and integrate relevant information scattered on different websites and social platforms.
Another application scenario is the oracle, where AI can obtain prices from multiple sources to provide accurate pricing data. For example, Upshot uses AI to target the volatile prices of NFTs, providing NFT prices with a percentage error of 3-10% through more than 100 million evaluations per hour.
▎AI and Development & Audit
Recently, a Web2 AI code editor called Cursor has attracted a lot of attention in the developer circle. On its platform, users only need to use natural language descriptions, and Cursor can automatically generate corresponding HTML, CSS and JavaScript codes, greatly simplifying the software development process. This logic is also applicable to improving the development efficiency of Web3.
Currently, deploying smart contracts and DApps on public chains usually requires following exclusive development languages such as Solidity, Rust, Move, etc. The vision of new development languages is to expand the design space of decentralized blockchains and make them more suitable for DApp development, but given the large gap in Web3 developers, developer education has always been a more difficult problem.
At present, the conceivable scenarios of AI assisting Web3 development include: automated code generation, smart contract verification and testing, DApp deployment and maintenance, intelligent code completion, AI dialogue to answer development questions, etc. The assistance of AI not only helps to improve development efficiency and accuracy, but also lowers the threshold for programming, allowing non-programmers to turn their ideas into practical applications, bringing new vitality to the development of decentralized technology.
Currently, the most eye-catching are one-click token launch platforms such as Clanker, an AI-driven "Token Bot" designed for fast DIY token deployment. You only need to tag Clanker on the SocialFi protocol Farcaster client such as Warpcast or Supercast, tell it your token idea, and it will launch the token for you on the public chain Base.
There are also contract development platforms, such as Spectral, which provides one-click generation and deployment of smart contracts to lower the threshold for Web3 development. Even novice users can compile and deploy smart contracts.
In terms of auditing, the Web3 auditing platform Fuzzland uses AI to help auditors check for code vulnerabilities and provide natural language explanations to assist audit expertise. Fuzzland also uses AI to provide natural language explanations of formal specifications and contract codes, as well as some sample code to help developers understand potential problems in the code.
3. AI and the New Narrative of Web3
The rise of generative AI brings new possibilities for new Web3 narratives.
NFT: AI injects creativity into generative NFTs. Through AI technology, a variety of unique and diverse artworks and characters can be generated. These generative NFTs can become characters, props or scene elements in games, virtual worlds or metaverse. For example, Bicasso under Binance allows users to generate NFTs by uploading pictures and entering keywords for AI calculations. Similar projects include Solvo, Nicho, IgmnAI, and CharacterGPT.
GameFi: Focusing on AI's natural language generation, image generation, and intelligent NPC capabilities, GameFi is expected to improve efficiency and innovation in game content production. For example, Binaryx's first blockchain game AI Hero allows players to randomly explore different plot options through AI; similarly, there is also a virtual companion game Sleepless AI, which is based on AIGC and LLM, and players can unlock personalized gameplay through different interactions.
DAO: Currently, AI is also envisioned to be applied to DAOs to help track community interactions, record contributions, reward members with the most contributions, proxy voting, etc. For example, ai16z uses AI Agent to collect market information on and off the chain, analyze community consensus, and make investment decisions based on the suggestions of DAO members.
Part.3 The significance of the combination of AI and Web3: towers and squares
In the heart of Florence, Italy, lies the most important venue for local political activities and a gathering place for citizens and tourists - the Central Square. Here stands a 95-meter-high Town Hall Tower. The vertical and horizontal visual contrast between the tower and the square complement each other, creating a dramatic aesthetic effect. Inspired by this, Niall Ferguson, a professor of history at Harvard University, associated it with the world history of networks and hierarchies in his book "The Square and the Tower", the two of which have risen and fallen over time.
This wonderful metaphor is not out of place when applied to the relationship between AI and Web3 today. From the long-term, non-linear history of the relationship between the two, we can see that squares are more likely to produce new things and be more creative than towers, but towers still have their legitimacy and strong vitality.
With the ability of technology companies to highly cluster energy, computing power and data, AI has exploded with unprecedented imagination. Major technology companies have invested heavily in the market. From different chatbots to the "underlying big model" GPT-4, GP4-4o and other iterative versions have appeared one after another, the automatic programming robot (Devin) and Sora with the initial ability to simulate the real physical world have come into being, and so on. The imagination of AI has been infinitely magnified.
At the same time, AI is essentially a scaled and centralized industry. This technological revolution has pushed technology companies, which have gradually mastered structural dominance in the "Internet era", to a narrower point. Huge electricity, monopoly cash flow, and the huge data sets required to dominate the intelligent era have created higher barriers for them.
As towers get taller, the number of decision-makers behind the scenes shrinks. AI centralization brings many hidden dangers. How can the people gathered in the square avoid the shadows under the towers? This is exactly the problem that Web3 hopes to solve.
Essentially, the inherent properties of blockchain enhance AI systems and bring new possibilities, mainly:
"Code is law" in the era of artificial intelligence - through smart contracts and cryptographic verification, transparent systems automatically execute rules and deliver rewards to people who are closer to the goal.
Token economics — creating and coordinating the actions of participants through token mechanisms, staking, slashing, token rewards, and penalties.
Decentralized governance – prompting us to question the sources of information and encouraging a more critical and discerning approach to AI technologies, preventing bias, misinformation, and manipulation, ultimately fostering a more informed and empowered society.
The development of AI has also brought new vitality to Web3. Perhaps the impact of Web3 on AI will take time to prove, but the impact of AI on Web3 is immediate: this can be seen from the Meme carnival and the AI Agent helping on-chain applications to lower the threshold for use.
When Web3 is defined as the self-entertainment of a small group of people and is caught in doubts about the replication of traditional industries, the addition of AI has brought it a foreseeable future: a more stable and larger Web2 user group, and more innovative business models and services.
We live in a world where "towers and squares" coexist. Although AI and Web3 have different timelines and starting points, their end point is how to make machines better serve humans. No one can define a rushing river. We look forward to seeing the future of AI+Web3.
*All content on the Coinspire platform is for reference only and does not constitute an offer or recommendation of any investment strategy. Any personal decision made based on the content of this article is the responsibility of the investor, and Coinspire is not responsible for any gains or losses arising therefrom. Investment is risky and decisions should be made with caution.
