Yesterday, the DeAI training platform Flock.io in the Web3 AI field officially announced a collaboration with Qwen, a large language model under Alibaba Cloud. If I remember correctly, this should be the first time web2 AI has actively initiated an integrated collaboration with web3 AI. Not only has Flock achieved a meaningful breakthrough, but it has also boosted the morale of the web3 AI track, which has been under pressure. Let me elaborate on this:
1) I have explained in the pinned tweet that previously, web3 AI Agents tried to stimulate the landing of Agent applications through tokenomics and engaged in a fast deployment competitive paradigm. However, after a wave of asset issuance FOMO, everyone realized that web3 AI had almost no chance of winning in practicality and innovation compared to web2 AI.
Therefore, the emergence of web2 innovative AI technologies such as Manus, MCP, and A2A has directly or indirectly burst the bubble existing in the Web3 AI Agent market, leading to a bloodbath in the secondary market.
2) How to break the deadlock? The path is quite clear. Web3 AI urgently needs to find an ecological niche that complements web2 AI to address high computing costs, data privacy issues, and the fine-tuning of models for vertical scenarios that centralized web2 AI cannot solve.
The reasons are straightforward: purely centralized AI models will inevitably face concentrated problems related to computing resource acquisition channels and costs, as well as data privacy issues. Meanwhile, the distributed architecture attempted by web3 AI can utilize idle computing resources to reduce costs, and will also protect privacy based on technologies like zero-knowledge proofs and TEE, while promoting model development and fine-tuning in vertical scenarios through data ownership and incentive contribution mechanisms. No matter how much criticism it faces, the decentralized architecture and flexible incentive mechanisms of web3 AI can have an immediate effect on solving some of the problems existing in web2 AI.
3) Speaking of the collaboration between Flock and Qwen, Qwen is an open-source large language model developed by Alibaba Cloud. Its outstanding performance in benchmark tests and the flexibility it offers developers for local deployment and fine-tuning have made it a common choice among some developers and research teams.
Flock is a decentralized AI training platform that integrates federated learning and distributed AI architecture. Its greatest feature is to protect user privacy through distributed training while keeping 'data local', ensuring transparency and traceability of data contributions, thereby addressing the issues of fine-tuning and application of AI models in vertical fields such as education and healthcare. Specifically, Flock has three key components:
1. AI Arena, this is a competitive model training platform where users can submit their models and compete for optimization results and rewards. Its main purpose is to incentivize users to continuously fine-tune and improve their local large models through a 'game-like' mechanism design, thereby selecting better benchmark models.
2. FL Alliance (Federated Learning Alliance), to address the cross-organizational collaboration issues in sensitive vertical scenarios such as traditional healthcare, education, and finance, the Federated Learning Alliance has achieved a way for multiple parties to enhance model performance together without sharing raw data through localized model training and a distributed collaboration framework.
3. Moonbase, it serves as the neural hub of the Flock ecosystem, equivalent to a decentralized model management and optimization platform, providing various fine-tuning tools and computing power support (computing power providers, data annotators). It not only offers a distributed model repository but also integrates fine-tuning tools, computing resources, and data annotation support to empower users to efficiently optimize local models.
4) So, how should we view the collaboration between Qwen and Flock? Personally, I believe the extended significance of their collaboration is even greater than the substance of the current cooperation.
On one hand, against the backdrop of continuous technological pressure from web2 AI on web3 AI, Qwen, representing the tech giant Alibaba, has gained a certain level of authority and influence within the AI circle. Qwen's active choice to collaborate with a web3 AI platform fully demonstrates the credibility of web2.
The recognition of AI by the Flock technical team, along with a series of research and development efforts between the Flock team and the Qwen team, will deepen the linkage between web3 AI and web2 AI.
On the other hand, the previous web3 AI was often just a shell of tokenomics and performed poorly in actual utility implementation. Although it explored various directions such as AI agents, AI platforms, and even AI frameworks, it failed to produce truly effective solutions in areas like DeFi and GameFi. This revelation from a web2 tech giant has, to some extent, set the tone for the future development path and focus of web3 AI.
The most critical point is that after experiencing a pure FOMO craze of 'asset issuance', web3 AI needs to regroup and focus on a goal that can deliver real results. In fact, web3 AI has never merely been a channel for easier and more efficient deployment of AI agents to issue assets, nor is it a game for raising funds through asset issuance. It needs to strive for cooperation with web2 AI to complement each other's ecological niches, truly demonstrating the indispensability of web3 AI in the current wave of AI trends.
I am very pleased to see more cross-border collaborations like web2 AI and web3 AI being achieved.