Introduction
AI development has made tremendous breakthroughs in recent years, driven by advancements in data, computational power, and algorithm research, particularly with the emergence of OpenAI GPT-4, which represents the arrival of foundational LLM large models, boosting productivity and transforming social efficiency.
However, the downsides represented by these closed-source large models like GPT-4 have also become apparent, as centralized models often face limitations in third-party integrations, impacting the scalability and interoperability of AI agents based on centralized models.
Therefore, open-source large models like Llama and others have garnered increasing attention from researchers, but open-source does not equate to transparency, as it faces many challenges.
The main dilemma is that open-source AI development offers no economic incentives for most contributors. Even though some competition rewards do exist, they are usually one-off, and subsequent improvement and development work still need to be powered by love, unless at a certain scale, with a large community of followers, more income realization possibilities will arise, encouraging more contributors to continue improving.
Thus, the Bittensor AI project attempts to utilize Web3 token mining to make open-source AI development more sustainable, verifiable, and efficient. Through Yuma Consensus, it aims to introduce and align resources among research parties (miners), validation parties (validators), and AI project parties (subnet creators), making AI research more transparent and decentralized, allowing anyone to contribute to AI and earn the rewards they deserve.
The performance of tokens in the secondary market also confirms people's expectations, with prices rising from over $50 in September 2023 to over $500 in December 2024, a tenfold increase!
Recently, the investor of Bittensor and founder of Digital Currency Group established an accelerator called Yuma dedicated to incubating subnet projects within the Bittensor ecosystem and serves as CEO, demonstrating his confidence and potential in the Bittensor project.
Image source: Coindesk
Of course, the success of any project cannot be achieved without scrutiny. Since the inception of Bittensor, there has been a lot of FUD. In this article, we summarize many unanswered questions and attempt to understand Bittensor's future positioning and potential in the decentralized AI race through research and analysis.
What is Bittensor?
Bittensor was founded in 2021 by a team from Toronto, Canada, consisting of Jacob Robert Steeves, Ala Shaabana, and Garrett Oetken.
Bittensor is a decentralized AI infrastructure used by AI developers to build and deploy machine learning models or other AI-related developments. Many Web3 AI projects, regardless of whether they already have their own blockchain, can connect to Bittensor's blockchain 'subtensor' and become part of a subnet.
What is a Subnet?
Subnets form the core of the Bittensor ecosystem, with each subnet being an independent incentive-based competitive market. Anyone can create a subnet, customize the tasks that subnet should perform, and design incentive mechanisms (in machine learning analogies, the incentive mechanism can be understood as a target loss function that guides model training towards ideal results). One only needs to pay a registration fee (in TAO) to create a subnet and receive a subnet's netuid. Note that a subnet creator does not need to undertake the operational tasks within the subnet but transfers the rights to operate those tasks to others.
Operating tasks for the subnet provide another way for others to participate, namely joining an existing subnet. If joining an existing subnet, there are two ways to participate: as a subnet miner or subnet validator. Aside from paying the registration fee priced in TAO (if a validator, they also need to stake TAO), one only needs to provide a computer with sufficient computational resources, register that computer and their wallet to a subnet, and run the subnet miner module or subnet validator module provided by the subnet creator (both modules are Python code in the Bittensor API).
How does the competitive market of subnets work?
The operation of subnet competition is as follows: suppose you decide to become a subnet miner. The subnet validator will assign some tasks for you to complete. Other miners in the subnet will also receive the same type of tasks. After all subnet miners complete their tasks, they will submit the results to the subnet validator.
Subsequently, subnet validators will independently evaluate and rank the quality of tasks submitted by subnet miners. As a subnet miner, you will receive rewards (priced in TAO) based on the quality of your work. Similarly, other subnet miners will receive corresponding rewards based on their performance. At the same time, subnet validators will also receive rewards for ensuring that high-quality subnet miners receive better rewards, thus promoting the continuous improvement of the overall quality of the subnet. All these competitive processes are automated based on the incentive mechanisms coded by the subnet creators.
Image source: Steps on how Subnet Creator defines Incentive Mechanism
The incentive mechanism ultimately serves as a judgment of the performance of subnet miners. When the incentive mechanism is well-calibrated, a virtuous cycle can be formed, and subnet miners continuously improve the tasks required in competition.
Conversely, poorly designed incentive mechanisms may lead to exploitations and shortcuts, adversely affecting the overall quality of the subnet and undermining the enthusiasm of honest miners.
The specific work of each subnet miner depends on the purpose the subnet creator had in mind when establishing that subnet, which can be more variable or more specific. For example, the task for miner in subnet 1 is to respond to the text prompts sent by the subnet validators and provide the best prompt completion results, or the task for miners in subnet 47 is to provide storage.
Each subnet also has its unique research and commercialization direction, such as attempting to tackle technical challenges in a specific AI industry, like decentralized AI training, verifiable inference, or providing foundational infrastructure and resources needed for AI, such as a GPU trading market or data annotation services. Alternatively, it may help users identify AIGC deepfake technology’s subnets, such as Subnet 34 - BitMind.
Currently, Bittensor has over 55 subnets, and this number continues to grow!
Image source: IOSG Ventures
Image source: IOSG Ventures
The role of the Subtensor blockchain
Clearly, blockchain and the project token TAO have played a significant role in this series of competitions.
Firstly, the Subtensor blockchain records all key activities of subnets in its ledger. More importantly, the Subtensor blockchain is responsible for determining the reward distribution among subnet miners and subnet validators. An algorithm called Yuma Consensus (YC) continuously runs on the Subtensor blockchain. Each subnet validator ranks the work quality of all subnet miners, and all rankings from each subnet validator are collectively sent as input to the YC algorithm. Generally, rankings from different subnet validators arrive at Subtensor at different times, but the YC algorithm on Subtensor waits for all rankings to arrive. Usually, every 12 seconds, the YC algorithm calculates rewards based on all validators' ranking inputs. These rewards (priced in TAO) will be deposited into the wallets of subnet miners and subnet validators. The Subtensor blockchain continuously runs the YC algorithm independently for each subnet.
The YC consensus algorithm mainly considers two factors: the first is a weight vector maintained by each subnet validator, where each element of the vector represents the weight assigned to the subnet miners, indicating the ranking of all subnet miners by each subnet validator based on the historical execution of that subnet validator. The second factor is the amount staked by each validator and miner. The on-chain Yuma consensus will use this weight vector and the amount staked to calculate rewards and distribute them among subnet validators and subnet miners.
The Bittensor API serves to transmit and connect the opinions of validators on the subnet with the Yuma consensus on the Subtensor blockchain. Additionally, validators within the same subnet only connect to miners within that subnet; validators and miners from different subnets do not communicate or connect with each other.
Image source: Bittenso
Game theory of Validators
To participate as a subnet validator or subnet miner, one must first register and stake. Registration means registering a key in the desired subnet, obtaining a UID slot in that subnet, which represents the right to validate that subnet. Note that subnet validators can simultaneously hold multiple UID slots and validate multiple subnets, but do not need to increase their amount staked; staking once is sufficient to choose multiple UID slots for validation across multiple subnets (similar to the concept of restaking).
Therefore, to obtain the maximum reward, staking validators tend to opt to provide validation services for all subnets. However, not all staked validators have the right to provide staking services. Only the top 64 validators ranked by the amount staked in a subnet are considered to have genuine validation permissions for that subnet. This reduces the risk of malicious actions by validators, as the amount staked becomes a high barrier and increases the cost of wrongdoing (one must stake at least 1,000 TAO to set weights in a subnet). Each validator will try to build a good reputation and performance record to attract more TAO delegated staking and increase their staked amount, aiming to become one of the top 64 validators in that subnet.
Once subnet validators and subnet miners (who do not need to stake) register their keys to the subnet, they can start mining.
Unique token incentive economy
All TAO token rewards are newly minted, similar to Bitcoin. Bittensor's $TAO has the same tokenomics and issuance curve as Bitcoin. TAO supply: a total cap of 21 million, halving every 4 years.
The network started fairly, with no pre-mined TAO tokens or ICO. Currently, the network generates 7,200 TAO per day, with 1 TAO generated per block, approximately every 12 seconds. The total supply cap is set at 21 million, following a programmed issuance plan similar to Bitcoin.
Image source: Bittensor
However, Bittensor introduces a unique mechanism whereby once half of the total supply is distributed, the issuance rate is halved. This halving occurs approximately every 4 years and continues at each half point of the remaining tokens until all 21 million TAO tokens are in circulation.
Although TAO adopts the issuance curve and philosophy of Bitcoin, due to its recovery mechanism, this curve is actively dynamic and not fixed like Bitcoin.
Recovery Mechanism:
The current daily token issuance for this cycle is 7,200 TAO (the same as Bitcoin's issuance during its first cycle from January 2009 to November 2012).
However, a certain amount of dynamic TAO is recovered daily through key (re)registrations.
To become a miner or validator, one must register a key in the network and meet other GPU and computational power requirements. Registration requires recovering TAO, that is, paying a certain amount of TAO to reinvest in the network.
Each time a key (re)registration occurs, that TAO is removed from the circulating supply and placed back into the protocol's issuance pool, where it can be mined again in the future.
This mechanism postpones the planned 4-year halving time, as the recovered TAO is dynamic. When more keys are (re)registered, the cost of recovering TAO increases, or other subnets are launched, the recovered TAO may significantly increase.
Moreover, registration applies not only to new entrants but also to those users whose registration has been canceled for the following reasons:
For miners, their models and inference are not competitive enough among other miners;
For validators, they fail to continuously set the correct weights, maintain issuance, or do not have enough TAO in their keys (self-delegation + market share of other delegators).
These factors will also exacerbate the growth of registration demand.
Recovered TAO quantity = Total number of registered (or re-registered) keys across subnets * average registration (or re-registration) cost
Therefore, the originally planned first halving after 4 years might be delayed to 5 or 6 years, or even longer. This entirely depends on the balance between the issuance and recovery of TAO.
The Bittensor network went live on January 3, 2021. According to token recovery data from taostats, the planned halving date is expected to be delayed to November 2025.
Image source: https://taostats.io/tokenomics
What is dTAO?
dTAO is an innovative incentive mechanism proposed by the Opentensor/Bittensor network, aimed at addressing the inefficiencies in resource allocation within decentralized networks. Unlike traditional methods where validators manually vote on resource allocation, dTAO introduces a market-based dynamic adjustment mechanism that directly ties resource allocation to the performance of subnet networks, thereby optimizing the fairness and efficiency of reward distribution.
Core Mechanism
Market-based dynamic resource allocation
The dynamic allocation mechanism of TAO is built upon the market performance of subnet tokens. Each subnet in the network has independent tokens, with their relative prices determining the distribution ratio of TAO issuance among subnets. As market information changes, this distribution ratio will dynamically adjust, ensuring resources flow to efficient and promising subnets.
Embedded liquidity pool design
Each subnet is configured with a liquidity pool consisting of TAO and subnet tokens (subnet/TAO token pairs). Users can exchange subnet tokens by staking TAO into the liquidity pool. This design incentivizes users to invest in high-performing subnets, indirectly supporting the overall development of the network.
Fair Token Distribution Mechanism
Subnet tokens are gradually distributed through a 'Fair Launch' model, ensuring that the team needs to earn token market share through long-term contributions and building. This mechanism avoids the risk of tokens being rapidly sold off while encouraging the team to focus on technological improvements and ecosystem building.
Balancing the roles of users and validators
The resource allocation of dynamic TAO is not only determined by the market but also jointly influenced by validators and users. Validators need to rigorously assess the technical capabilities, market potential, and actual performance of the teams, similar to venture capitalists (VC). Users drive the market value formation of subnets further through staking TAO and participating in market transactions.
Economic model analysis
Current funding support
Data shows that the subnets in the network currently average about $47,000 in rewards per day, corresponding to an annual support of about $17 million. This funding scale is far higher than the median seed round (about $3 million) and Series A financing (about $14 million) for traditional AI startups, providing strong support for the rapid development of subnets.
Future potential
Currently, Bittensor's annual budget is projected to reach $1.3 billion, comparable to centralized AI research institutions like OpenAI and Anthropic. With the introduction of dynamic TAO, future issuances of TAO will primarily flow to the liquidity pools of subnet tokens, further promoting the circulation of capital and value within the ecosystem.
Long-term incentives
The design of dTAO greatly incentivizes teams to continually improve their technology and applications by linking issuance to market performance. This mechanism also curbs short-term behaviors that rapidly cash out through over-the-counter (OTC) trades, laying the foundation for the network's long-term sustainable development.
Impact and significance
Resource allocation optimization
dTAO adjusts resource allocation through market dynamics, ensuring that subnets with high utility and high growth potential receive more resources. This mechanism not only improves overall network efficiency but also drives competition and innovation.
Decentralized AI ecosystem construction
Bittensor is not only a decentralized AI network but also serves as an incubation platform for AI networks through dynamic TAO. The competition and cooperation between subnets further promote the development of the decentralized AI ecosystem.
Incentives for ecosystem participants
Dynamic TAO balances the interests of users, validators, and teams, ensuring that all participants can contribute to the growth of the network through economic incentive mechanisms.
Enhancement of the validator role
Validators need to play a more significant role in the network. They rigorously assess the value and potential of subnets in a venture capital-like manner, ensuring the scientific and rational distribution of network resources.
The launch of dTAO symbolizes a significant advance in decentralized network resource allocation mechanisms. Through market-based dynamic adjustments, embedded liquidity pool designs, and fair issuance models, dTAO achieves efficient and equitable resource allocation. Furthermore, as an AI network incubation platform, it not only empowers the development of subnets but also provides new developmental pathways for the future of decentralized AI networks.
Applications of Agents on Bittensor
Many people say that Bittensor is the AI coin represented by VC elites and has fallen behind in the flourishing application era of various agents' developer architectures. With the recent rise of AI Agents and the total market value of AI Agent-related tokens surpassing $10 billion, particularly projects represented by the Virtuals ecosystem, which dominate a total market cap of $5 billion (including various practical investment and research analysis-type Agents like $AIXBT, $VADER, $SEKOIA, etc.), Bittensor seems to be left behind in the eyes of many.
However, in reality, Bittensor still possesses many 'Alphas.' Many people do not realize that the success of Virtuals/ai16z in the consumer AI Agent industry and the efforts of Bittensor subnets in decentralized AI infrastructure are complementary.
As the TVL (Total Value Locked) and influence of Agents expand, robust training and inference infrastructure become increasingly important.
Currently, Virtuals have collaborated extensively with Bittensor in the ecosystem.
Image source: Virtuals
Many consumer-oriented virtuals protocol agents are supported by Bittensor subnets, leveraging the computational power and data ecosystem of TAO to create new possibilities.
$TAOCAT
TAOCAT is an AI agent created by Masa within the Virtuals ecosystem, primarily serving as a staunch defender of TAO, actively participating in discussions on X and voicing for TAO's influence.
TAOCAT utilizes the real-time data infrastructure of subnet 42 Masa and the advanced LLM provided by Bittensor subnet 19 to compete for the distribution of TAO tokens in the Agent Arena of Bittensor subnet 59, creating a new paradigm for tokenized AI value capture, where any user's interaction on X will become training data for TAO Cat.
Other projects supported by Bittensor subnets include:
$AION: The first agent capable of predicting prediction outcomes and participating in prediction market betting, with a copy-trading feature set to launch soon.
$SERAPH: The first project focused on verification infrastructure, aimed at certifying the AI Agent wave that is about to sweep our digital world.
The collaboration between Virtuals and Bittensor demonstrates the immense practical value that can be created on the Bittensor infrastructure. With the official launch of AgenTAO (SN62), this will become a significant milestone for automated software engineering agents on Bittensor, as all Bittensor subnets will gradually be developed by agents on Bittensor. In the future, we will see more application-side AI agents emerging from the Bittensor ecosystem!
Image source: taogod
Conclusion
The future of Bittensor is exciting, with many research and investment institutions focusing on the Bittensor ecosystem emerging, similar to the Ethereum network. This includes the DCG founder’s involvement, podcasts, blogs, and OSS Capital, which focuses on research within the Bittensor ecosystem while also being a research organization of a subnet. A network of connections resembling a PayPal mafia for Bittensor has formed, with recent gatherings among Contango, Canonical, Delphi Labs, and DCG, where many experts from the Crypto x AI field have also started to gravitate towards and support Bittensor. Therefore, it is not unreasonable that Bittensor was able to surpass Virtuals in Kaito's mindshare recently.
Image source: BitMind Bittensor Subnet 34
In April next year, 2025, Austin, Texas, Bittensor will host The Endgame Summit, a conference and hackathon focused on introducing more subnets, validators, and miners into the Bittensor ecosystem and expanding its territory.
Image source: Endgame Summit
Regardless of whether it is a centralized AI project or a decentralized AI project, the final standard will return to the product itself. Currently, Bittensor's ecosystem has emerged and is flourishing.
Image source: Outpost AI Research
Recently, the founder of Bittensor summarized the major achievements of various subnets on Bittensor over the past year on his personal X account:
Image source: Xhttps://x.com/const_reborn/status/1873359385373909008
Therefore, let us continue to look forward to Bittensor and see what products and applications will emerge from Bittensor in the future, becoming the go-to place for people seeking specific AI solutions!
This article is authorized for reposting from: (ForesightNews)
Original author: IOSG Ventures
'A Comprehensive Introduction to Bittensor (TAO)! How to Mine with Tokens and Reshape the Future of AI Development' was first published on 'Crypto City.'
