Heurist is a Layer 2 network based on the ZK Stack that focuses on AI model hosting and inference. It aims to decentralize AI using blockchain technology to promote accessibility and unbiased innovation. Similar to the web3 version of HuggingFace, Heurist provides users with serverless access to open-source AI models hosted on a decentralized computing resource network.
The name Heurist comes from “heuristics,” which refers to the process by which the human brain quickly reaches reasonable conclusions or solutions when solving complex problems. This name reflects Heurist’s vision of using decentralized technology to quickly and efficiently solve AI model hosting and inference problems.
Closed-source AI models often comply with US laws and regulations, which may not align with the needs of other countries and cultures, leading to over-censorship or under-representation. This not only affects the performance of AI models but also infringes on users’ freedom of expression.
On the other hand, open-source AI models have shown superior performance in many areas compared to closed-source models. For example, Stable Diffusion models have outperformed OpenAI’s DALL-E 2 in image generation and have lower costs. Open-source models provide transparency, allowing developers and artists to fine-tune them according to specific needs.
The community-driven innovation of open-source AI is another highlight. Open-source AI projects benefit from collective contributions and reviews from diverse communities, promoting rapid innovation and improvement. Open-source AI models offer unprecedented transparency, allowing users to review training data and model weights, enhancing trust and security.
To address data privacy concerns, Heurist integrates the Lit Protocol for encrypted data transmission during AI model inference, including input and output. Heurist has two types of miners: public miners and privacy-enabled miners. Public miners process data without encryption, while trusted node operators can become privacy-enabled miners and handle sensitive information such as confidential files, health records, and user identity data. Only miners that match the user’s access control conditions (ACC) can decrypt the data.
To establish trust in privacy-enabled miners, Heurist uses off-chain consensus established through real-world laws or agreements, which is technically feasible. Additionally, trusted execution environments (TEE) can be used to ensure the security and confidentiality of sensitive data. Although mature TEE solutions for large AI models are not yet available, recent advancements in chips from companies like Nvidia have shown potential in supporting TEE for AI workloads.
Heurist has its own utility token called HUE, with a dynamic supply regulated through issuance and burning mechanisms. The maximum supply of HUE tokens is limited to 1 billion. Users can mine HUE tokens by hosting AI models on their GPUs, and mining rewards are in the form of esHUE tokens, which are automatically compounded into the miner’s stake. The mining reward rate depends on GPU efficiency, availability, the type of AI model being run, and the total stake in the node.
Users can also stake HUE or esHUE tokens in miner nodes and receive rewards in HUE or esHUE tokens. Unstaking HUE tokens requires a 30-day lock-up period, while unstaking esHUE tokens has no lock-up period. esHUE rewards can be converted into HUE tokens through a one-year linear release period. Users can transfer their staked HUE or esHUE tokens from one miner node to another, promoting flexibility and competition among miners.
Similar to Ethereum’s EIP-1559 model, Heurist implements a token burning mechanism. When users pay for AI inference fees, a portion of the HUE payment is permanently removed from circulation. The issuance and burning of tokens are balanced based on network activity, and during periods of high usage, the burning rate may exceed the issuance rate, leading to a deflationary phase for the Heurist network. This mechanism helps regulate token supply and align token value with actual demand within the network.
Heurist also incorporates a bribery mechanism, inspired by Curve Finance, to enhance mining efficiency. Miners can set a certain percentage of their mining rewards as bribes to attract stakers. Stakers may choose miners offering the highest bribes, considering factors such as hardware performance and uptime. Miners are incentivized to provide bribes because higher stakes lead to higher mining efficiency, creating a competitive and cooperative environment for miners and stakers to provide better services to the network.
During the incentivized testnet phase, Heurist allocates 5% of the total token supply to mining rewards. These rewards are in the form of points that can be converted into fully liquid HUE tokens after the mainnet token generation event (TGE). The testnet rewards are divided into two categories: one for Stable Diffusion models and another for Large Language Models (LLMs).
For LLM miners, they earn Llama Points for processing input/output tokens from the Mixtral 8-7b model. For Stable Diffusion miners, they earn Waifu Points for generating 512×512-pixel images using the Stable Diffusion 1.5 model. The allocation ratio of Llama Points and Waifu Points will be determined closer to the TGE, considering the demand and usage of both model categories in the coming months.
Participation in the testnet can be done through two methods: hosting AI models on your own GPU or renting hosted mining nodes from Heurist. Supported GPUs for mining are listed on the Heurist website. The testnet has anti-cheat measures in place, with an asynchronous monitoring system storing and tracking the inputs and outputs of each computational task. Heurist reserves the right to reduce testnet points for miners engaging in malicious behavior to manipulate the reward system.
Heurist showcases its powerful AI capabilities and wide range of applications through image generation, chatbots, and AI search engines. Their image generation application relies on the Stable Diffusion model to generate high-quality images based on text prompts. The chatbot application utilizes large language models for intelligent conversations and content generation. The AI search engine combines pre-trained language models to provide accurate information retrieval and detailed answers.
These applications not only enhance user experience but also demonstrate Heurist’s innovation and technical advantages in decentralized AI. The effectiveness of these applications can be seen in the provided images. Users can interact with these models through API requests, submitting text inputs and receiving model-generated responses, enabling diverse conversations and interactions.
In conclusion, Heurist aims to create a dynamic and efficient token economy to support its decentralized AI model hosting and inference network. While DePIN (Decentralized Protocol for Intelligent Networks) shows great potential for development and growth, it still faces challenges in terms of technological maturity, service stability, market acceptance, and regulatory environment. However, as technology advances and the market evolves, these challenges are expected to be gradually overcome. Once these challenges are effectively addressed, DePIN could experience massive adoption, bringing in a large number of new users and attention to the crypto domain, potentially becoming a catalyst for a new bull market. Let us witness the arrival of that day together!
