Written by: Pzai, Foresight News
In the future cyber society, AI is an important part of human interaction, and its infrastructure construction will naturally become a key breakthrough for development. The large amount of computing and reasoning service demand generated by model operation is also booming. As the AI-as-a-Service infrastructure in the Web3 field, the decentralized AI cloud computing protocol Heurist recently announced the completion of a $2 million financing round with participation from Amber Group, Contango Digital, Manifold Trading, Selini Capital and other institutions. What kind of excellent infrastructure will they bring to future crypto users and AI developers?
Infrastructure
With the booming development of generative AI, the global market size will reach US$356.1 billion by 2030. This will be accompanied by a rapid growth in demand for computing resources and reasoning services.
Today, many AI models run on traditional centralized cloud services, but accessing these services has the following problems:
Centralization risk: In centralized cloud services, all data and computing are concentrated in a single data center, which is vulnerable to single point failure. In addition, all data is stored in a single location, increasing the risk of data leakage and abuse.
High cost: Running AI models requires a lot of computing resources, especially when training large models, which consumes a lot of GPU and CPU resources. These resources are very expensive, especially for companies that need to continuously run and update models, which brings high long-term operating costs.
Low resource utilization: In traditional centralized cloud services, the uneven load of AI model training and reasoning leads to insufficient resources during busy hours and waste of resources during idle hours. In addition, due to possible resource grabbing between services, resources cannot be fully utilized, further reducing resource utilization.
As an AI-as-a-service infrastructure, Heurist allows GPU owners to provide computing resources without permission through the DePIN protocol architecture, and opens APIs to allow users and developers to access and run AI models based on these computing resources.
Computational Layer
The computing layer consists of computing nodes, Pods, and a verification system. Computing nodes serve as the basic unit, allowing computing providers (miners) to join in a permissionless and indefinite manner.
General pattern of Pods
Pod is an independently deployable workload unit using GPU. Through Pod, users can specify certain computing requirements to match the corresponding node. It supports automatic load adjustment and can be used in AI model operation, fine-tuning and ZK proof generation. At the same time, Heurist uses crypto-economic verification to ensure the integrity and reliability of the calculation results.
Orchestration Layer
Heurist builds ZK Stack to provide high throughput and low-cost environment, while providing users with sovereign isolation of interoperability to ensure the stability of operations and resource management. In addition, for the on-chain sequencer, transaction calculations will be executed off-chain first, and then on-chain after successful calculation and execution. This execution mode can reduce the execution cost of failed transactions.
Application Layer
Any user can build a front-end on the application layer, and Heurist provides two payment gateway modes, in which developers can either prepay a certain amount of resources in the form of computing points, or pay directly according to user usage by distributing the costs to users. The latter can also provide autonomous payment options for AI Agents.
In summary, the advantage of Heurist lies in its decentralized architecture, which improves the system's fault resistance and data security. At the same time, it reduces operating costs by utilizing idle computing resources, improves resource utilization through dynamic resource allocation, and enhances the flexibility and scalability of the system through load regulation. The crypto-economic verification built by ZK Stack ensures the integrity and reliability of the calculation results, enabling the entire system to efficiently and securely adapt to various workload requirements and provide a stable experience for users and developers.
Products and Applications
Mainstream applications built on Heurist today include Heurist Imagine (AI NFT art creation), Pondera (AI dialogue), Heurist Search (AI search), and Babel Bot (Discord translation robot), covering mainstream AI use cases. With decentralized GPU computing, the overall cost efficiency is many times higher than that of closed-source models. In addition to common raw image models, Heurist also provides fine-tuning models focusing on Web3 Agents and IPs, such as the Theia language model developed by Chainbase, the ZKsync ecosystem-native IP Zeek model, and the Magus Devon model created by the Forgotten Runes community. Through these highly customized Web3 AI models, Heurist drives a variety of AI Agents built on the blockchain, which are composable and censorship-resistant. With on-chain GPU computing, the overall cost efficiency is many times higher than that of closed-source models, making it the model layer and computing power layer of the AI Agent ecosystem.
Ecological Construction
In terms of economic system construction, in the future, Heurist's native token HEU can pay for computing resources for developers, serve as the gas cost of user transactions, and support pledge and governance functions. In terms of share distribution, the emission scale for computing power providers is 50%, and the early community distribution (including early miners and NFT holders) accounts for 7%. Heurist's elastic token release rules dynamically adjust the annualized inflation rate between 1.25% and 5%, realizing the elastic expansion and contraction of computing power resources to adapt to changing computing needs.
In its early stages of construction, Heurist issued a community Heurist Imaginaries NFT, a series of 500 AI-generated artworks. The airdrop plan includes an airdrop of 1.75% of HEU tokens to holders of this series.
Heurist has established partnerships with partners in multiple fields. In terms of GPU computing power, Heurist has cooperated with companies such as Aethir, Exabits, Spheron, and Mintair to provide powerful computing resources, which provide a foundation for the training and inference network construction of AI models. In the field of AI and data infrastructure, Heurist has established partnerships with GenLayer, Chasm Network, Aizel, and Gateway Network to ensure the efficient flow and security of data.
In terms of consumer applications, Heurist has worked with companies such as CreatorBid, Arbus AI, DecentAI, and Eden Art to bring AI technology to a wider consumer market. In the field of AI Agents, Heurist provides customized raw image models for CreatorBid, and in particular, customized services for the WallStreetBets brand. Through these collaborations, the Heurist platform is continuously expanding its influence and promoting the application and development of AI technology in various fields.
According to the roadmap, the project will issue tokens in the fourth quarter of 2024, issue the first quarter mining token airdrops and launch the developer API payment function, launch the staking function in 2025, and gradually launch subsequent mining and ZK chains, etc., while launching user payment functions and validator systems, etc. Users can participate by contributing GPUs (12 GB or more of video memory, such as GTX 3090 or 4090), participating in Intract points activities, participating in Create-to-earn activities in Heurist Imagine, and minting AI NFTs to earn ZK tokens.
Conclusion
The future cyber society will rely more on the interaction between AI and humans, and Heurist solves the centralization risks, high costs and low resource utilization of traditional centralized cloud services, and also provides AI developers and users with a new way to utilize computing resources. It is expected to become a bridge connecting AI technology and the future, promote the development of society in a more intelligent direction, and provide solid technical support for building a more intelligent, interconnected and secure cyber society. Users can visit the project homepage, Discord, Twitter and Medium for further information.