Author: Haotian

Recently observing the AI industry, I have noticed an increasingly "downward" trend: evolving from the previous mainstream consensus of concentrated computing power and "large" models, a branch has emerged that leans towards local small models and edge computing.

This can be seen from Apple Intelligence covering 500 million devices, to Microsoft's launch of the Windows 11 dedicated 330 million parameter small model Mu, to Google's DeepMind's robot "offline" operations, and so on.

What will be different? Cloud AI competes on parameter scale and training data, with the ability to burn money as the core competitive advantage; local AI competes on engineering optimization and scene adaptation, making further progress in privacy protection, reliability, and practicality. (The illusion problem of major general models will severely affect the penetration of vertical scenarios.)

This actually presents a greater opportunity for web3 AI. When everyone was competing on "generalization" (computing, data, algorithms), they were naturally monopolized by traditional giant companies. Trying to compete with Google, AWS, OpenAI, etc., under the concept of decentralization is simply wishful thinking, as there are no resource advantages, technological advantages, and certainly no user base.

However, in a world of localized models + edge computing, the situation faced by blockchain technology services is vastly different.

When AI models run on user devices, how can we prove that the output results have not been tampered with? How can we achieve model collaboration while protecting privacy? These questions are precisely the strengths of blockchain technology…

I have noticed some new web3 AI-related projects, such as the recently launched data communication protocol Lattica by @GradientHQ, which received a $10M investment from Pantera, aimed at solving the data monopoly and black box issues of centralized AI platforms; @PublicAI's brainwave device HeadCap collects real human data to build a "human verification layer," already achieving $14M in revenue; in fact, they are all attempting to solve the "trustworthiness" issue of local AI.

In short: Only when AI truly "sinks" into every device will decentralized collaboration shift from concept to necessity?

Web3AI projects, instead of continuing to compete in the generalization track, should seriously consider how to provide infrastructure support for the wave of localized AI?