The Future Development Path of AI+Web3 (1): Industrial Picture and Narrative Logic

Preface

Over the past year, the advent of large generative AI models like ChatGPT has propelled AI beyond basic automation tools to sophisticated decision-making and predictive systems, establishing itself as a pivotal force driving societal advancement. This evolution has led to a surge in AI products and applications, with ChatGPT introducing notable products such as GPTs and Sora. NVIDIA, a key player in AI infrastructure, has consistently outperformed expectations, with its data center business contributing over 83% of revenue in the fourth quarter of fiscal year 2024, marking a remarkable 409% growth. Notably, 40% of this revenue was attributed to large model inference scenarios, underscoring the escalating demand for robust computing power infrastructure.

Currently, AI has garnered significant attention from European and American capital circles, while the Web3 market is experiencing a resurgence in the bull market. The convergence of AI and Web3 represents the intersection of two highly sought-after technical trends. Recent emergence of projects centered around this theme underscores the market’s keen interest and high expectations for this convergence.

Setting aside the hype and price bubbles, what is the current state of development in the AI+Web industry? Are there tangible application scenarios? Looking ahead, can we generate value, establish a narrative, and shape an industry? How will the AI+Web3 industry evolve in terms of ecological patterns, and what potential directions lie ahead?

Future3 Campus will delve into these topics through a series of articles, dissecting every facet of the AI+Web3 industry chain. This inaugural piece will provide an overview of the industry landscape and narrative framework of AI+Web3.

AI work production process

In essence, the integration of AI+Web3 unfolds in two key dimensions. Firstly, the role of Web3 in advancing AI development, and secondly, the fusion of Web3 applications with AI technology. Presently, the predominant focus lies on the empowerment of AI through Web3 technology and concepts across various projects. Thus, the analysis can delve into the integration of AI with Web3, spanning from model training to production. The advent of LLM introduces some variances from traditional machine learning processes, yet a streamlined AI production process typically encompasses the following stages:

1 data collection

Data serves as the cornerstone throughout the AI model training lifecycle, necessitating high-quality datasets for exploratory data analysis (EDA) to establish reproducible, editable, and shareable data sets, tables, and visualizations.

2 Data preprocessing and feature engineering/hint engineering

Post data acquisition, preprocessing is imperative, involving feature engineering in machine learning and prompt engineering in large models. This entails iterative classification, aggregation, and deduplication to identify intricate features, alongside the iterative development of prompts for structured queries in LLM. Moreover, ensuring reliable storage and sharing of features/prompts is essential.

3 Model training and tuning

Leveraging a diverse model library to train AI models, enhancing performance, efficiency, and accuracy through ongoing iteration and adjustment. Notably, in LLM, continuous model tuning is achieved via human feedback reinforcement learning (RLHF).

4 Model review and governance

Utilizing the MLOps/LLMOps platform to optimize model development processes encompassing model discovery, tracking, sharing, and collaboration. This ensures model quality and transparency while adhering to ethical and compliance standards.

5 Model reasoning

Deploying trained AI models to make predictions on novel, unseen data by utilizing learned parameters to process input data and generate prediction outcomes such as classification or regression predictions.

6 Model deployment and monitoring

Following performance validation against set standards, deploying the model into real-world applications and instituting continuous monitoring and maintenance to uphold optimal performance amidst evolving environments.

Throughout this process, numerous opportunities exist for integrating Web3 technologies. Presently, challenges in AI development like model transparency, bias, and ethical application have garnered widespread attention. In this context, the amalgamation of Web3 technology with cryptographic solutions like ZK can enhance trust in AI systems. Furthermore, the escalating demand for AI applications underscores the need for cost-effective and open infrastructure and data networks. Web3’s distributed network and incentive models can foster more open-source AI networks and communities.

AI+Web3 industrial landscape and narrative logic

By integrating the AI production process outlined above with the integration of AI with Web3 and examining the prevalent AI+Web3 projects in the current market, we have delineated the landscape of the AI+Web3 industry. This industry chain can be segmented into three layers: the infrastructure layer, middle layer, and application layer.

1. Infrastructure Layer

The infrastructure layer primarily encompasses computing and storage infrastructure essential for the entire AI workflow and production process. It facilitates the computing power necessary for AI model training, inference, and data storage throughout the lifecycle.

The rapid expansion of AI applications has spurred a surge in demand for infrastructure, particularly high-performance computing capabilities. Hence, the provision of enhanced performance, cost-efficiency, and robust computing and storage infrastructure is anticipated to emerge as a pivotal trend in the early stages of AI development, capturing over 50% of the industry chain value.

Web3 technology has the potential to establish a decentralized computing and storage resource network, leveraging idle and dispersed resources to substantially reduce infrastructure costs and cater to a broad spectrum of AI application requirements. Consequently, decentralized AI infrastructure stands as the prevailing narrative.

Notable projects in this domain include Render Network, focusing on rendering services, and Akash and Gensyn offering decentralized cloud services and computing hardware networks. In the storage sector, prominent projects such as Filecoin and Arweave have recently introduced storage and computing services tailored for the AI field.

1. Middle Layer:

The middle layer primarily involves leveraging Web3 technologies to address current challenges and enhance processes within AI production. Key aspects include:

1) Data Acquisition Stage: Utilizing decentralized data identity to establish an open data network/trading platform, safeguarding user privacy and data integrity through cryptography and blockchain features. This approach incentivizes users to share high-quality data, expanding data sources and enhancing data acquisition efficiency. Notable projects in this realm include Worldcoin and Aspecta for AI identity, Ocean Protocol for data trading, and Grass for a low-participation threshold data network.

2) Data Preprocessing Stage: Establishing a distributed AI data annotation and processing platform, employing economic incentives to encourage crowdsourcing models for efficient and cost-effective data preprocessing, benefiting subsequent model training. Representative projects include Public AI.

3) Model Verification and Inference Stage: Addressing the challenges of opaque data and model processes in AI, Web3 technologies like ZK and homomorphic encryption can be integrated to verify model reasoning, ensuring the accuracy of models while safeguarding input data privacy. A typical application scenario is ZKML. Representative projects combining Web3 technology in model verification and inference stages include Bittensor, Privasea, and Modulus.

Projects in the middle layer often prioritize developer tools, offering supplementary services to developers and project stakeholders. Market demand and commercial implementation of AI are still evolving in the early stages of development.

3 Application layer

At the application level, the focus shifts to the utilization of AI technology within Web3 contexts. The integration of Web3 applications with AI technology aims to enhance efficiency and elevate product experiences. AI functionalities such as content generation, analysis, and prediction find diverse applications in gaming, social networking, data analysis, and financial forecasting. Presently, AI+Web3 applications can be categorized into three main types:

1) AIGC Type: Leveraging AI generative technology to enable users to create text, images, videos, avatars, and other content through interactive dialogue. This AI functionality can be presented as a standalone AI agent or seamlessly integrated into products. Notable projects in this category include NFPrompt and SleeplessAI.

2) AI Analysis Category: Project teams leverage proprietary data, knowledge bases, and analytical capabilities to train vertical AI models for analysis, decision-making, and prediction tasks. These AI models are offered as products to users, granting access to AI analysis capabilities for tasks like data analysis, information tracking, code auditing, financial forecasting, and more. Representative projects include Kaito and Dune.

3) AI Agent Hub: Serving as a hub for various AI agents, this category typically enables users to create customized AI agents without coding, akin to GPTs. Noteworthy projects in this realm include My Shell and Fetch.ai.

While there are no standout projects yet in the application layer, it holds significant growth potential in the long term. Success in the realm of AI+Web3 applications hinges not solely on technological innovation but on the accumulation of product capabilities and technical expertise. Particularly in the AI domain, products offering superior user experiences will gain a competitive edge in this evolving landscape.

**Disclaimer:

1. This article is reprinted from [marsbit], the original title is “The future development path of AI+Web3 (1): Industrial landscape and narrative logic”, the copyright belongs to the original author [Wanxiang Blockchain], if you have any objection to the reprint, please contact Gate Learn Team, the team will handle it as soon as possible according to relevant procedures.

2. Disclaimer: The views and opinions expressed in this article represent only the author’s personal views and do not constitute any investment advice.

3. Other language versions of the article are translated by the Gate Learn team, not mentioned in Gate.io, the translated article may not be reproduced, distributed or plagiarized.