Beyond Bittensor: Are These Crypto AI Networks Worth Your Attention?

Over the past year, the combination of decentralized AI and various AI tools has turned AI + Web3 into a hot topic within the cryptocurrency community. Currently, over 140 projects merge Web3 with AI, spanning computation, verification, the metaverse, and gaming. Ethereum co-founder Vitalik Buterin has explored use cases for integrating blockchain and AI, highlighting an increase in meaningful and robust applications. This topic was also prominently featured at the recent Hong Kong Web3 Carnival.

This article highlights three Web3 and AI projects, exploring their unique positions and potential in the crypto AI field.

Bittensor: Despite Leading in Market Value, Faces Questions about its Practical Utility.

In the AI sector, where work typically revolves around technologically intensive tasks rather than resource-intensive ones, a major issue has been the lack of effective cooperation among algorithms and models due to technical barriers, resulting in a zero-sum situation. Bittensor proposes a solution through a blockchain network and incentive mechanisms to encourage cooperation among different algorithms, gradually creating a shared knowledge market akin to bitcoin mining, but for training and validating AI models.

The name “Bittensor” combines “Bit” and “Tensor.” “Bit” is well-known from Bitcoin as the smallest unit of currency, and in broader computing terms, represents the most basic unit of information. “Tensor,” derived from the Latin “Tendera” meaning “to extend,” refers in physics to a multidimensional array or matrix representing various data types. In the field of machine learning, it signifies entities that handle multidimensional data.

Bittensor’s architecture is dual-layered: the foundational layer is a blockchain built on Polkadot Substrate that manages consensus and incentives; the AI layer handles reasoning and training, ensuring compatibility across Bittensor protocol nodes. Key participants in the Bittensor network are miners and validators. Miners submit training models for token rewards, while validators ensure the validity and accuracy of model outputs. Bittensor employs the Yuma consensus mechanism, which combines PoW and PoS elements, to distribute incentives. Miners compete to earn token rewards through computation, and validators stake tokens on a subnet and verify outputs to earn TAO incentives—the more accurate and consistent their evaluations, the greater the rewards.

Subnets form the backbone of the Bittensor ecosystem. In October 2023, the “Revolution” upgrade introduced the concept of “Subnet,” allowing different subnets to manage various tasks such as machine translation and image recognition, and to interact and learn from each other. Anyone can create a subnet on Bittensor by paying with TAO tokens, the amount of which depends on the demand for subnets on the network. Before a subnet can go live, it must undergo testing on local and test networks.

Currently, Bittensor hosts one special subnet, #0 Root, and 32 others. Subnet #0 Root, established by the Opentensor Foundation, serves as Bittensor’s governance hub, distributing TAO generated to other subnets. On Subnet #0 Root, validators are the top 64 stakers from other subnets, while the role of miners is played by other subnets. Additionally, #0 Root allocates incentives based on contributions. For the remaining 32 subnets, validators and miners receive shares of TAO based on their contributions: 41% each to validators and miners, with the remaining 18% going to subnet creators. Competition among subnets is fierce; the system caps the number of allowed subnets at 32, although over 200 subnets are waiting for mainnet registration in the testnet. Recently, notable teams like MyShell TTS have registered their subnets on Bittensor. According to subnet registration rules, once the cap is reached, the subnet with the lowest token allocation is automatically deregistered.

Bittensor has recently been criticized for its high registration fees and questioned practical utility. Currently, the cost to register a subnet on Bittensor is 2,078.49 TAO, which soared to 10,281 TAO on March 1st—equivalent to over 7 million USD. As the price of TAO increases, these fees are expected to rise further. Each new project registration doubles the fee, and if no new registrations occur, the price is halved linearly over four days. Such steep costs can be a significant burden for developers wishing to create or join subnets. Additionally, doubts about the utility of Bittensor’s subnets have surfaced. Most of the 32 subnets cater to low-barrier applications like “data sorting” and “text, image, and audio conversion.” Teams building on Bittensor typically include no more than a dozen full-time members, with many having only two to three people. Critics like Eric Wall, founder of the Bitcoin Ordinals project Taproot Wizards and Bitcoin NFT project Quantum Cats, have dismissed Bittensor as a pointless decentralized experiment that offers no real utility. Wall described Subnet #1 as merely a text-prompt service where miners respond to prompts similarly to ChatGPT, with validators only checking for answer similarity—denying rewards to any miner who deviates, which severely limits efficiency and fails to verify actual model operations. He argues that the network serves no purpose beyond operating internally and is essentially a way to purchase useless AI tokens for the sake of decentralized AI visibility.

Ritual: Boosted by a Stellar Background, Using ZKP for AI Model Reasoning Training

The existing AI stack faces numerous challenges, including the lack of safeguards for computational integrity, privacy, and censorship resistance. Moreover, infrastructure dominated by a few centralized companies restricts developers’ and users’ ability to integrate locally, leading to issues with effectiveness. Against this backdrop, Ritual, a decentralized AI computing platform, has been launched.

Ritual’s main goal is to provide an open and modular sovereign execution layer for AI, exploring how AI can be integrated into environments like EVM, SVM, and other virtual machines. In essence, Ritual connects a network of distributed computing resources with model creators, allowing creators to host their AI models while enabling users to add comprehensive, verifiable reasoning capabilities of these models to their workflows.

Ritual boasts an impressive team. Its co-founders, Niraj Pant and Akilesh Pott were previously general partners at Polychain. The team also includes senior engineers from major companies such as Microsoft AI and Facebook Novi, as well as professionals from renowned institutions like Dragonfly, Protocol Labs, and dYdX. Furthermore, Ritual’s advisory panel is notable, including EigenLayer founder Sreeram Kannan, Gauntlet CEO Tarun Chitra, and BitMEX co-founder Arthur Hayes among others.

To date, Ritual has completed two funding rounds. In November 2023, Ritual announced a $25 million funding round led by Archetype, with participation from accomplice, Robot Ventures, dao5, Accel, Dilectic, Anagram, Avra, and Hypersphere, as well as angel investors such as Coinbase’s former CTO Balaji Srinivasan, Protocol Labs researcher Nicola Greco, Worldcoin engineer DC Builder, EigenLayer’s Chief Strategy Officer Calvin Liu, Monad co-founder Keone Hon, and AI+Crypto project Modulus Labs’ Daniel Shorr and Ryan Cao. Then, on April 8, 2024, Ritual received a multimillion-dollar investment from Polychain Capital, although the exact amount remains undisclosed.

Ritual has launched Infernet, a lightweight library that integrates computing on the blockchain. This enables smart contract developers to request off-chain computations through Infernet nodes and deliver results to on-chain smart contracts using the Infernet SDK. Infernet nodes, which are light clients operating off-chain, primarily listen for on-chain or off-chain requests and deliver workflow outputs and optional proofs through on-chain transactions or off-chain APIs. The Infernet SDK is a set of smart contracts that allow users to subscribe to the outputs of off-chain computational workloads. One of its primary uses is to bring machine learning inference to the blockchain. Infernet can be deployed on any chain, facilitating integration with any protocol or application. Additionally, it allows developers to incorporate their proof systems, including Halo2 and Plonky3 validators.

Infernet functions similarly to an Oracle system—it does not perform inference directly on the blockchain. Instead, it issues requests on-chain, which are executed off-chain by nodes, and the results are then returned to the chain. This method has an inherent asynchronous issue, where developers must wait on the blockchain after making a request, and cannot get immediate responses. Ritual’s approach allows developers to perform inference computations directly in their familiar environments without worrying about where the computations are happening. While these operations still occur off-chain, embedding these computational operations within virtual machines allows each node to execute highly optimized AI operations while running a modified virtual machine. This method, which can be seen as a form of inter-process communication achieved through precompilation, also reflects a growing trend in the blockchain ecosystem.

In practical terms, through Infernet, developers can delegate computationally intensive operations to off-chain nodes, which then use on-chain callbacks to utilize outputs and optional proofs in smart contracts, thus circumventing the limitations of the smart contract execution environment. For example, Emily is developing a new NFT collection that allows minters to independently add new features to their NFTs. She has set up a minting website that sends signed requests to Infernet nodes running custom workflows. These workflows parse user inputs to generate new images, and the Infernet nodes then send the final images back to her smart contract through on-chain transactions.

At the end of 2023, Ritual launched a chatbot named Frenrug in the Friend.tech chatroom, supported by the Infernet SDK. Users holding a Frenrug Key can send messages to Frenrug to buy or sell Friend.tech user Keys. However, Frenrug doesn’t process messages directly; instead, it forwards them to multiple Infernet nodes running different language models. These nodes process the messages and generate votes on the blockchain. Once enough votes are collected, the system aggregates these votes to execute the corresponding actions on the blockchain, such as buying or selling Keys. Frenrug then responds in the chatroom, displaying the voting results and final actions taken, providing transparency on how user requests are handled.

Ritual is developing its second product, “Ritual Chain,” a sovereign blockchain. While Infernet can be easily integrated into any EVM chain, allowing any protocol to utilize its capabilities, Ritual believes that building a dedicated chain is essential. This approach will enable more efficient features at the core execution and consensus layers, fulfilling the aspirations of users who want to maximize the value AI brings to their protocols. Establishing a sovereign chain will involve developing various types of validators, proof systems, and complex functionalities that are also user-friendly for easy adoption.

Virtual: More Engaging and User-Focused

Unlike Bittensor and Ritual, which interact with various machine models, Virtual Protocol is akin to a decentralized factory focused on creating AI characters for virtual worlds. It emphasizes user participation and integrates human subjectivity and social consensus into its vision, fostering personalization and immersion. Virtual Protocol’s core idea is that future virtual interactions will be powered by AI and built in a decentralized manner to offer personalized and immersive experiences. Personalization ensures each interaction makes a unique connection with users, enhancing relevance, while immersion can stimulate multiple senses to create a more realistic experience.

Participants in the Virtual ecosystem include contributors and validators. Contributors can supply text, voice, and visual data to improve existing models or propose new ones. This content is reviewed and certified by validators to ensure accuracy and authenticity, and to evaluate the quality of contributions according to the standards set by the Virtual Protocol ecosystem.

New proposals: Anyone can initiate the creation of a Genesis Virtual but must stake at least 100,000 VIRTUAL tokens within three months and follow the DAO proposal process. All token holders within the Virtual community can vote on these proposals. Once approved, a new Virtual NFT is minted.

Contributing to existing models: Proposals are automatically generated, reviewed, discussed, validated, and voted on by validators to decide whether to implement changes.

Currently, only validators are authorized to validate or vote on proposals, and the process is conducted anonymously. Validators must interact with each model at least ten times. Upon completing their tasks, validators receive stake rewards proportional to their total staked amount. To become a validator, a user must hold 1,000 Virtual tokens in their Virtual account and commit to validating all proposals. Additionally, Virtual employs a DPos mechanism, allowing users to delegate any amount of tokens to a validator to earn stake rewards without validating themselves, after which validators return the stake rewards to delegators minus a 10% income fee.

The entire participation process in Virtual is transparent and recorded on the public blockchain. All contributions are transformed into NFTs and stored in the Immutable Contribution Vault (ICV) to ensure traceability and fair reward distribution. The ICV is a multi-layered on-chain repository that archives all approved contributions on the blockchain, presenting the current status of each Virtual and tracking its historical evolution. Open-sourcing the VIRTUALs codebase model within the ICV creates a transparent environment that enhances composability, allowing developers and contributors to build on existing VIRTUALs and integrate seamlessly.

Virtual tokens are central to the Virtual Protocol, rewarding contributors and validators, supporting protocol development, and facilitating airdrops. Of the 1 billion total supply of Virtual tokens, 60% are in circulation, 5% are reserved for liquidity pools, and the remaining 35% are allocated for community incentives and initiatives to develop the Virtual Protocol ecosystem. Token release over the next three years will not exceed 10% annually and must be approved by the regulatory body before deployment.

Virtual Protocol drives growth through revenue and incentives derived from various dApp usages, which pay fees to the protocol. At the end of each month, Virtual Protocol distributes the incentives based on the total inflow of dApp revenue; 10% goes to the protocol, and the remaining 90% is distributed among various Virtual applications according to their stake ratio, ensuring that income is proportional to their contributions. For example, if the total income inflow is $100, $10 is allocated to the protocol. Of the remaining $90, if Virtual A has a staking pool with 9,000 tokens and Virtual B has 1,000 tokens, Virtual A would receive 90% of $90 = $81, while Virtual B would receive 10% of $90 = $9.

Within each Virtual application, revenue is evenly distributed between validators and contributors. Validators earn based on operational time and the amount staked, where operational time is the ratio of the number of proposals they verify to the total number of proposals. For instance, if Validator A in Virtual A operates 90% of the time, they would earn 81/2*90% = $36.45. This income is further distributed to all stakers according to their stakes, with a default of 10% serving as a fee paid to the pool’s validator. Contributors receive income based on utilization rates and impact pools, considering the duration their contributions are actively used in the system. Developers and maintainers of models receive 30% of the total allocated income, while those providing and maintaining datasets for model fine-tuning receive 70%. Impact pools award points based on the significance of contributions, facilitating a fair distribution of rewards.

Virtual has integrated its AI capabilities into a game called AI Waifu, set in the fictional world of Arcadia. In this game, players assume the role of wizards who engage in battles with other wizards and their Waifus. Players can deepen their bond with their Waifu by engaging in dialogues to unlock hidden stories and earn rewards by giving gifts. The game offers three different Waifus to choose from, each with its own unique backstory and personality traits. Additionally, the game features a combat mode where players can attempt to charm other Waifus and protect their own. All in-game expenditures contribute to the game’s reward pool, with 60% of WAI transaction fees also being allocated to this pool to distribute as prizes.

Unlike other AI companions and chatbots, AI Waifu is visually represented as a 3D model and can react to voice and text with emotional animations. Through interactions, AI Waifu continuously learns and provides personalized responses, avoiding repetitive content styles. Additionally, AI Waifu is a cross-platform Progressive Web App (PWA) with a crypto-enabled economic design, allowing for co-ownership and returning part of its expenditures as revenue shares to developers. Beyond AI Waifu, Virtual plans to launch a new AI RPG with cross-game memory and ultimate consciousness. These AI agents can evolve dynamically through interactions within the game, retain memories from one game when transferred to another, and mimic human behaviour, enhancing the player’s immersive experience. They also allow users to upload interaction logs to earn tokens and plan to introduce a virtual idol that can livestream across platforms.

Conclusion

In the realm of crypto AI, Bittensor, Ritual, and Virtual Protocol each carve out their niches. Bittensor is focused on creating a market for shared algorithmic knowledge and currently leads in market capitalization within the crypto AI sector. However, its subnet registration fees and practical utility have raised concerns among community members recently. Whether the issues of a single subnet can be attributed to the entire network remains under review. Additionally, in response to problems related to the system’s heavy reliance on validators, the Opentensor Foundation contributors have proposed a dynamic TAO solution, “BIT001”.

Ritual, backed by a robust financing lineup and team credentials, has emerged as a rising star in the crypto AI space. Previously, Dragonfly partner Haseeb Qureshi noted that Ritual’s cryptoeconomic approach is the simplest and potentially cheapest in the verified inference space, although it carries risks of node collusion. However, the founders of Ritual have clarified on social media that their platform does not use a cryptoeconomic model based on node collaboration and selective collusion, instead offering users options to choose their desired level of security.

In contrast, Virtual Protocol is noted for its engaging nature and emphasis on user participation. For instance, it has launched the AI Waifu virtual companion game and plans to introduce a game AI agent that dynamically evolves based on player interactions and the gaming environment. This aims to enhance the social and continuity aspects of gaming, making it more engaging compared to traditional games with fixed rules.

Beyond these three projects, there are numerous other crypto AI initiatives worth attention, such as io.net focusing on the GPU leasing market, the AI agent protocol Autonolas, and the creator-centric Web3 enabled AI platform MyShell. These projects showcase the diversity and potential of the crypto AI sector, which continues to evolve rapidly.

Disclaimer：

1. This article is republished from panewslab, originally titled “Beyond Bittensor: Are These Crypto AI Networks Worth Your Attention?” All copyrights belong to the original author, ChainFeeds. If you have any objections to this repost, please contact the Gate Learn team.
2. The views and opinions expressed in this article are solely those of the author and do not constitute investment advice.
3. Other language versions of this article have been translated by the Gate Learn team and may not be copied, disseminated, or plagiarized without mentioning Gate.io.