Decentralized Machine Learning Networks in the Wave of the AI Revolution

With the rapid development of artificial intelligence technology, we are entering a data-driven new era. Breakthroughs in fields such as deep learning and natural language processing have made AI applications ubiquitous. The launch of ChatGPT in 2022 sparked explosive growth in the AI industry, followed by a series of AI tools such as text-to-video and automated office applications. The market value of the AI industry has also rapidly increased, expected to reach $185 billion by 2030.

However, the current AI industry is mainly dominated by a few tech giants, and this concentration of technology has also brought challenges such as data monopolies and uneven distribution of computing resources. The decentralization concept of Web3 offers new possibilities for addressing these issues and is expected to reshape the AI development landscape.

Amidst this wave of AI, many high-quality Web3+AI projects have emerged. Among them, the Bittensor project stands out, building an AI algorithm platform with a built-in filtering mechanism through blockchain competition and incentive mechanisms, aiming to retain the best AI projects.

Bittensor: Pioneer of Decentralization in Machine Learning Networks

Bittensor is a decentralized incentive machine learning network and digital goods marketplace. It operates on a distributed computer network controlled by different entities, employing a fair incentive mechanism to provide services to individuals in need of machine learning computing resources.

Unlike many highly valued venture capital projects, Bittensor is more like a fair, interesting, and meaningful technology-driven project. Its development process did not involve "making grand promises" or "deceiving investors:"

• In 2021, the project was created by a group of technical experts promoting Decentralization AI networks, using the Substrate framework to build the blockchain.
• In 2022, the Alpha version of the network was released to validate the feasibility of Decentralization AI. The Yuma consensus was introduced, emphasizing the principle of data unawareness.
• In 2023, the Beta version was released, introducing a token economic model (TAO) to incentivize network maintenance.
• In 2024, apply DHT technology to improve data storage and retrieval efficiency, and begin promoting the concepts of subnet and digital goods market.

The Bittensor token TAO is similar to Bitcoin in many ways, with a total supply of 21 million coins and a halving every four years. TAO is distributed through a fair launch, with no pre-mining or allocations reserved for the team and investors. Currently, a block is generated every 12 seconds, with each block rewarding 1 TAO, resulting in an approximate daily output of 7200 TAO, distributed among various subnets and their participants based on contribution.

As of now, the total number of accounts on the Bittensor network has exceeded 100,000, with nearly 80,000 non-zero accounts. In the past year, the TAO price has seen a maximum increase of several dozen times, with the current market value at approximately 2.278 billion USD and the coin price at 321 USD.

The Core of Bittensor: Subnet Architecture

The Bittensor protocol is a decentralized machine learning protocol that allows network participants to exchange machine learning capabilities and predictions, facilitating shared collaboration of models and services in a peer-to-peer manner. The network consists of multiple subnetworks and employs a survival of the fittest mechanism, where underperforming subnetworks are replaced by new ones.

Subnets can be seen as independently running code snippets that establish specific user incentives and functionalities, while maintaining the same consensus interface as the main network. Currently, there are 45 subnets besides the root subnet, and this is expected to increase to 64 between May and July 2024.

The subnet mainly includes three types of roles:

1. Subnet Owner: Provides the underlying code, sets incentive mechanisms, and allocates miner rewards.

2. Miners: Run servers and mining code to stay ahead through competition. A miner can run nodes on multiple subnets.

3. Validator: Measure subnet contributions and verify correctness to earn rewards. You can stake TAO for additional income.

Subnetwork emission is the reward distribution mechanism in the Bittensor network, typically allocating 18% to owners, 41% to validators, and 41% to miners. The subnetwork consists of 256 slots, with 64 allocated to validators and 192 to miners. The performance of validators and miners determines their status and rewards.

After the subnet registration, there is a 7-day immunity period, and the first registration fee is 100 TAO. When all positions are filled, the new subnet will replace the one with the lowest emission that is not in the immunity period. Therefore, the subnet needs to continuously increase its staking amount and efficiency to avoid being eliminated.

Innovative Consensus and Proof Mechanisms

The Bittensor network employs various consensus and proof mechanisms, among which the most distinctive are the PoI( proof of intelligence) mechanism and the Yuma consensus.

The PoI mechanism verifies participant contributions through intelligent computation of tasks, ensuring network security and efficient resource utilization. Miners complete tasks assigned by validators, and validators score based on the quality of completion.

Yuma consensus is the core consensus mechanism. After the validator scoring input algorithm, validators with more stakes have higher scoring weights. The algorithm will exclude abnormal results and ultimately distribute rewards based on the comprehensive score. This mechanism follows the principle of data unawareness, protecting privacy and security.

In addition, Bittensor introduces the MOE( mixture of experts) mechanism, integrating multiple expert-level sub-models to collaboratively address problems in different domains. Validators can score and rank the expert models and allocate rewards to incentivize continuous optimization.

Subnet Project Ecosystem

Currently, Bittensor has 45 registered subnets, 40 of which have been named. As the quota increases, registering new subnets becomes easier, but it also faces more intense competition. In the long run, subnets with poorer performance will be eliminated.

The top three subnets are:

• Vision 19: Focus on Decentralization image generation and inference, providing access to open source LLM and image models.
• Cortex.t on the 18th: Build a cutting-edge AI platform that provides text and image responses through APIs.
• Text Generation Subnet 1: Bittensor's first subnet project, focusing on text generation.

Other subnets also include different types such as data processing and trading AI. In terms of risk and return, a successful operation for more than a few weeks can yield considerable profits, but new nodes require high-performance graphics cards and optimized algorithms to survive in the competition.

Future Outlook

1. The Web3+AI sector is expected to maintain market enthusiasm in the long term, attracting substantial investment.

2. The Bittensor project architecture combines technological innovation and market recognition, with significant growth since its launch.

3. Its innovative subnet architecture lowers the threshold for AI teams to migrate to a Decentralization network, and the competitive mechanism also promotes continuous optimization.

4. As the number of subnets increases, the TAO rewards obtained from the original subnet may decrease, and it is necessary to pay attention to the changes in earnings.

5. The increase in the number of subnets may also bring about the problem of uneven project quality, and it is necessary to be cautious of inferior projects.