In the previous surge of decentralized AI, standout projects like Bittensor, io.net, and Olas quickly became industry leaders thanks to their cutting-edge technologies and visionary strategies. However, as these projects’ valuations soar, entry barriers for regular investors have also risen. Amid the current sector rotation, are there still fresh opportunities for involvement?

Flock: A Decentralized AI Training and Verification Network

Flock is a decentralized AI model training and application platform that combines federated learning with blockchain technology, offering users a secure environment for model training and management while safeguarding data privacy and facilitating fair community participation. The term “Flock” first gained prominence in 2022 when its founding team released an academic paper titled “FLock: Defending malicious behaviors in federated learning with blockchain.” The paper proposed using blockchain to combat malicious actions in federated learning. It outlined how a decentralized approach can bolster data security and privacy during model training, showcasing the potential applications of this innovative architecture in distributed computing.

Following initial concept validation, Flock launched the decentralized multi-Agent AI network, Flock Research, in 2023. In Flock Research, each Agent is a large language model (LLM) fine-tuned for specific domains, capable of providing users with insights across various fields through collaboration. In mid-May 2024, Flock officially opened the testnet for its decentralized AI training platform, allowing users to participate in model training and fine-tuning using the test token FML and earn rewards. As of September 30, 2024, the number of daily active AI engineers on the Flock platform has surpassed 300, with over 15,000 models submitted in total.

As the project continues to grow, Flock has also attracted attention from the capital markets. In March of this year, Flock completed a $6 million funding round led by Lightspeed Faction and Tagus Capital, with participation from DCG, OKX Ventures, Inception Capital, and Volt Capital. Notably, Flock is the only AI infrastructure project to receive a grant in the 2024 Ethereum Foundation’s academic funding round.

The Cornerstone of Reshaping AI Production Relationships: Introducing Smart Contracts for Federated Learning

Federated Learning is a machine learning approach that allows multiple entities (often referred to as clients) to collaboratively train models while ensuring that data remains stored locally. Unlike traditional machine learning, federated learning avoids uploading all data to a central server, thereby protecting user privacy through local computation. This method has already been applied in various real-world scenarios; for instance, Google introduced federated learning into its Gboard keyboard in 2017 to optimize input suggestions and text predictions while ensuring that user input data is not uploaded. Tesla also employs similar technology in its autonomous driving system, enhancing the vehicle’s environmental perception locally and reducing the need for massive video data transmission.

However, these applications still face challenges, particularly regarding privacy and security. Firstly, users need to trust centralized third parties. Secondly, during the transmission and aggregation of model parameters, it is crucial to prevent malicious nodes from uploading false data or harmful parameters, which could lead to biases in the overall model performance or even erroneous predictions. Research conducted by the FLock team, published in the IEEE journal, indicates that the accuracy of traditional federated learning models drops to 96.3% when 10% of the nodes are malicious, and further declines to 80.1% and 70.9% when the proportions of malicious nodes increase to 30% and 40%, respectively.

To address these issues, Flock introduced smart contracts on the blockchain as a “trust engine” within its federated learning framework. As a trust engine, smart contracts can automate the collection and validation of parameters in a decentralized environment, allowing for unbiased publication of model results and effectively preventing malicious nodes from tampering with data. Compared to traditional federated learning solutions, FLock’s model accuracy remains above 95.5%, even with 40% of the nodes being malicious.

The AI Execution Layer: Analyzing FLock’s Three-Layer Architecture

The key problem in the current AI landscape is that resources for AI model training and data usage remain highly concentrated among a few large companies, making it difficult for ordinary developers and users to effectively utilize these resources. Consequently, users are left with pre-built standardized models and can’t customize them according to their specific needs. This mismatch between supply and demand leads to a situation where, despite abundant computing power and data reserves in the market, they cannot be transformed into practically usable models and applications.

To tackle this issue, Flock aims to serve as an effective scheduling system that coordinates demand, resources, computational power, and data. Drawing on the Web3 technology stack, Flock positions itself as the “execution layer,” primarily responsible for allocating users’ customized AI requirements to various decentralized nodes for training, using smart contracts to orchestrate these tasks across global nodes.

Additionally, to ensure fairness and efficiency throughout the ecosystem, the FLock system is also responsible for the “settlement layer” and “consensus layer.” Settlement layer refers to incentivizing and managing participants’ contributions, rewarding or penalizing them based on task completion. Consensus layer involves assessing and optimizing the quality of training results, ensuring that the final generated models represent the global optimal solution.

The overall product architecture of FLock comprises three major modules: AI Arena, FL Alliance, and AI Marketplace. The AI Arena is responsible for decentralized foundational model training, FL Alliance focuses on model fine-tuning under the smart contract mechanism, and AI Marketplace serves as the final model application market.

AI Arena: Incentives for Localized Model Training and Validation

AI Arena is Flock’s decentralized AI training platform where users can participate by staking Flock testnet tokens (FML) and receive corresponding staking rewards. Once users define the models they need and submit tasks, training nodes within the AI Arena will train the models locally using the specified initial model architecture, without requiring direct data uploads to centralized servers. After each node completes training, validators are responsible for assessing the work of the training nodes, checking the quality of the models and scoring them. Those who do not wish to participate in the validation process can delegate their tokens to validators for rewards.

Within the AI Arena, the reward mechanisms for all roles depend on two core factors: the amount of tokens staked and the quality of the tasks. The staked amount reflects the participants’ “commitment,” while task quality measures their contribution. For example, the rewards for training nodes depend on the amount staked and the ranking of the submitted model quality, while validators’ rewards hinge on the consistency of voting results with consensus, the number of staked tokens, and the frequency and success rate of their participation in validations. The returns for delegators depend on the validators they choose and the amount they stake.

AI Arena supports traditional machine learning model training modes, allowing users to choose to train on local data from their devices or publicly available data to maximize the performance of the final model. Currently, the AI Arena public testnet has a total of 496 active training nodes, 871 validation nodes, and 72 delegators. The platform’s staking ratio stands at 97.74%, with average monthly earnings of 40.57% for training nodes and 24.70% for validation nodes.

FL Alliance: A Fine-Tuning Platform Managed by Smart Contracts

The highest-rated models on AI Arena are selected as “consensus models” and assigned to FL Alliance for further fine-tuning. This fine-tuning process consists of multiple rounds. At the beginning of each round, the system automatically creates an FL smart contract related to the task, which manages task execution and rewards. Similarly, each participant is required to stake a certain amount of FML tokens. Participants are randomly assigned roles as either proposers or voters. Proposers use their local datasets to train the model and upload the trained model parameters or weights to other participants. Voters then summarize and vote to evaluate the proposer’s model update results.

All results are submitted to the smart contract, which compares the scores from each round with those from the previous round to assess improvements or declines in model performance. If the performance score improves, the system advances to the next stage of training; if it declines, the training will restart using the previously validated model for another round of training, summarization, and evaluation.

FL Alliance achieves the goal of collaboratively training a global model with multiple participants while ensuring data sovereignty by combining federated learning and smart contract mechanisms. By integrating different data sources and aggregating weights, it can build a global model that performs better and possesses greater capabilities. Additionally, participants demonstrate their commitment to participation by staking tokens and receive rewards based on model quality and consensus results, forming a fair and transparent ecosystem.

AI Marketplace: How to Provide Personalized AI Services?

The models trained in AI Arena and fine-tuned in FL Alliance will ultimately be deployed in the AI Marketplace for use by other applications. Unlike traditional “model marketplaces,” AI Marketplace not only offers ready-made models but also allows users to modify these models and integrate new data sources to address different application scenarios. Furthermore, the AI Marketplace incorporates Retrieval-Augmented Generation (RAG) technology to enhance the accuracy of models in specific domains. RAG is a method that augments large language models by retrieving relevant information from external knowledge bases during response generation, ensuring that the model’s responses are more accurate and personalized.

Currently, the AI Marketplace has launched many customized GPT models based on different application scenarios, including BTC GPT, Farcaster GPT, Scroll GPT, and Ethereum GPT. Let’s take BTC GPT as an example to illustrate the difference between customized models and general models.

In December 2023, when asked “What is ARC20?” simultaneously to BTC GPT and ChatGPT:

• BTC GPT responded: “The ARC20 protocol is built on the Bitcoin network and allows users to create, transfer, and upgrade digital items, such as NFTs and fungible tokens. With the ARC20 protocol, non-fungible NFTs and fungible tokens can be generated on the Bitcoin network. The main token of the ARC20 protocol is ATOM, which is used for various purposes within the protocol.”
• ChatGPT replied: “As of my last knowledge update in January 2022, there was no widely recognized standard or technology known as ARC20 in the fields of technology, finance, or other common areas. It is possible that a new technology or standard called ARC20 has emerged since then, but I do not have relevant information.”

From their responses, we can see the importance and advantages of customized GPT models. Unlike general-purpose language models, customized GPT models can be trained on data specific to certain fields, thereby providing more accurate responses.

What Are the Similarities and Differences Between Flock and Bittensor, Both Supported by DCG?

As the AI sector revives, Bittensor, one of the representatives of decentralized AI projects, has seen its token rise by over 93.7% in the past 30 days, reaching near its historical peak, with its market cap surpassing $4 billion once again. Notably, Flock’s investment firm, Digital Currency Group (DCG), is also one of the largest validators and miners in the Bittensor ecosystem. According to sources, DCG holds approximately $100 million in TAO, and in a 2021 article by “Business Insider,” DCG investor Matthew Beck recommended Bittensor as one of the 53 most promising crypto startups.

Despite both being projects supported by DCG, Flock and Bittensor focus on different aspects. Specifically, Bittensor aims to build a decentralized AI internet, using “subnets” as its basic unit, where each subnet represents a decentralized market. Participants can join as “miners” or “validators.” Currently, the Bittensor ecosystem comprises 49 subnets, covering various domains such as text-to-speech, content generation, and fine-tuning large language models.

Since last year, Bittensor has been a focal point in the market. On one hand, its token price has skyrocketed, soaring from $80 in October 2023 to a peak of $730 this year. On the other hand, it has faced various criticisms, including questions about the sustainability of its model, which relies on token incentives to attract developers. Additionally, the top three validators in the Bittensor ecosystem (Opentensor Foundation, Taostats & Corcel, and Foundry) collectively hold nearly 40% of the staked TAO, raising user concerns about the level of decentralization.

In contrast, Flock aims to provide personalized AI services by integrating blockchain into federated learning. Flock positions itself as the “Uber of the AI space,” serving as a “decentralized scheduling system” that matches AI needs with developers. Through on-chain smart contracts, Flock automatically manages task allocation, result validation, and reward settlement, ensuring that each participant can fairly receive a share based on their contributions. Similar to Bittensor, Flock also offers users the option to participate as delegates.

Specifically, Flock provides the following roles:

• Training Nodes: These nodes can participate in AI task training competitions by staking tokens, suitable for users with computational capabilities and AI development experience.
• Validators: They must also stake tokens to join the network, responsible for verifying the quality of miners’ models and influencing reward distribution through submitted validation scores.
• Delegators: They can delegate tokens to miner and validator nodes to enhance the weight of those nodes in task allocation while sharing in the rewards earned by the delegated nodes. This allows users without technical expertise in training or validating tasks to participate in the network and earn rewards.

Flock.io has officially opened the delegation feature, allowing any user to stake FML tokens to earn rewards. Users can choose the optimal nodes based on their expected annualized returns to maximize their staking rewards. Flock also indicates that staking and related operations during the testnet phase will affect potential airdrop rewards after the mainnet launch.

In the future, Flock aims to introduce a more user-friendly task initiation mechanism that allows individuals without AI expertise to easily engage in creating and training AI models, realizing the vision of “everyone can participate in AI.” Flock is also actively pursuing various collaborations, such as developing an on-chain credit scoring model with Request Finance and partnering with Morpheus and Ritual to create trading bot models that offer one-click deployment templates for training nodes, making it simple for developers to start and run model training on Akash. Additionally, Flock has trained a Move language programming assistant to support developers on the Aptos platform.

Overall, while Bittensor and Flock have different market positions, both seek to redefine production relationships within the AI ecosystem through distinct decentralized technologies. Their shared goal is to dismantle the monopoly of centralized giants over AI resources and foster a more open and equitable AI ecosystem, which is urgently needed in today’s market.

Disclaimer：

1. This article is reprinted from [chainfeeds], All copyrights belong to the original author [Linda Bell]. If there are objections to this reprint, please contact the Gate Learn team, and they will handle it promptly.
2. Liability Disclaimer: The views and opinions expressed in this article are solely those of the author and do not constitute any investment advice.
3. Translations of the article into other languages are done by the Gate Learn team. Unless mentioned, copying, distributing, or plagiarizing the translated articles is prohibited.