Development Directions and Challenges of Web3 AI

NVIDIA's stock price has reached a new high, and the advancements in multimodal models have deepened the technological barriers of Web2 AI. From semantic alignment to visual understanding, from high-dimensional embedding to feature fusion, complex models are integrating various modalities of expression at an unprecedented speed, creating an increasingly closed AI stronghold. The US stock market has also expressed optimism with tangible actions, as both cryptocurrency-related stocks and AI stocks have experienced a small bull market.

However, this surge seems unrelated to the cryptocurrency sector. Recent attempts at Web3 AI, especially in the exploration of Agents, have shown significant deviations in direction. The attempt to assemble a Web2-style multimodal modular system with a decentralized structure is, in fact, a dual misalignment of technology and thinking. In today's environment, where module coupling is extremely strong, feature distribution is highly unstable, and computational power demands are increasingly centralized, multimodal modularity struggles to gain a foothold in the Web3 environment.

The future of Web3 AI lies not in imitation, but in strategic circumvention. From semantic alignment in high-dimensional space to information bottlenecks in attention mechanisms, and then to feature alignment under heterogeneous computing power, Web3 AI needs to adopt a tactical strategy of "surrounding the city from the countryside."

Web3 AI based on a flattened multimodal model, semantic alignment difficulties lead to poor performance

In modern Web2 AI's multimodal systems, "semantic alignment" refers to mapping information from different modalities into the same semantic space, allowing models to understand and compare the underlying meanings behind these distinctly different signals. Only under the premise of achieving a high-dimensional embedding space does it make sense to divide workflows into different modules for cost reduction and efficiency improvement. However, the Web3 Agent protocol cannot achieve high-dimensional embedding because modularity is essentially an illusion of Web3 AI.

Web3 AI requires the implementation of high-dimensional space, which essentially demands that the Agent protocol independently develop all involved API interfaces, contradicting its original intention of modularity. The high-dimensional architecture necessitates end-to-end unified training or collaborative optimization, while the "module as plugin" concept of Web3 Agent exacerbates fragmentation.

To achieve a full-chain intelligent agent with industry barriers, it requires end-to-end joint modeling, unified embedding across modules, and systematic engineering of collaborative training and deployment to make breakthroughs. However, there are currently no such pain points in the market, and naturally, there is no corresponding market demand.

In low-dimensional space, the attention mechanism is difficult to design precisely.

High-level multimodal models require the design of sophisticated attention mechanisms. The prerequisite for the attention mechanism to function is that the multimodal has high dimensionality. In high-dimensional space, a sophisticated attention mechanism can find the most core parts from a vast high-dimensional space in the shortest time.

Modular Web3 AI finds it difficult to achieve unified attention scheduling. First, the attention mechanism relies on a unified Query-Key-Value space, while independent APIs return data in different formats and distributions, lacking a unified embedding layer. Second, multi-head attention allows for simultaneous parallel focus on different information sources within the same layer, while independent APIs often involve linear calls, lacking the ability for parallel and dynamic multi-route weighting. Finally, a true attention mechanism dynamically allocates weights to each element based on the overall context, whereas in the API model, modules can only see the independent context at the time they are called.

Discrete modular assembly leads to feature fusion remaining at a superficial static stitching.

Feature fusion is the further combination of feature vectors obtained from different modalities after processing, based on alignment and attention, for direct use in downstream tasks. Web3 AI is certainly still at the simplest concatenation stage, as the premise for dynamic feature fusion is high-dimensional space and a sophisticated attention mechanism.

Web2 AI tends to use end-to-end joint training, while Web3 AI more often adopts a modular approach with discrete components. Web2 AI can calculate the importance scores of various features in real-time based on context and dynamically adjust the fusion strategy, while Web3 AI often fixes weights in advance or uses simple rules to determine whether to fuse.

Web2 AI maps all modal features into a high-dimensional space with thousands of dimensions, and the fusion process involves various high-order interactive operations. In contrast, the outputs of each agent in Web3 AI often consist of only a few key fields or metrics, with very low feature dimensions, making it difficult to express complex cross-modal associations.

Barriers in the AI industry are deepening, but pain points have yet to emerge.

The multimodal system of Web2 AI is an extremely large engineering project that requires massive amounts of data, powerful computing power, advanced algorithms, and excellent talent. This comprehensive, full-stack systematic work creates a strong industry barrier and forms the core competitiveness of a few leading teams.

Web3 AI or any cryptocurrency project aimed at product-market fit needs to adopt the "surrounding the cities from the countryside" tactic. It should start small-scale trials in edge scenarios to ensure a solid foundation before waiting for the emergence of core scenarios. The core of Web3 AI lies in decentralization, and its evolution path is reflected in high parallelism, low coupling, and compatibility of heterogeneous computing power.

Currently, the barriers of Web2 AI are just beginning to form, marking the early stage of competition among leading enterprises. Only when the dividends of Web2 AI have almost disappeared will the pain points it leaves behind become opportunities for Web3 AI to penetrate. Before that, we need to carefully identify protocols that have the potential for "surrounding cities from the countryside," paying attention to whether they can continuously iterate in small scenarios and whether they possess the flexibility to adapt to different situations. If the protocol itself relies too heavily on infrastructure and has a large network architecture, it is likely to face a high risk of elimination.