Roles & Functions
AI Oracle 2.0 is a mechanism to execute AI calling and validate AI processes. It plays a significant role in passing data from real-world to smart contracts.
Functions of AI Oracle 2.0 include:
- Defining inter-operational metadata standards for Oraichain and other blockchains
- Validating AI processes
- Providing trustworthy proofs
- Maintaining decentralization of AI functions
From Genesis AI Oracle to AI Oracle 2.0
Originally, Oraichain focused on solving the trustworthiness of AI to provide decentralization for Web3 dApps. In the original design, AI executor nodes play two roles in verification and aggregation.
Difficulties of this design are:
- Testing speed: calling multiple tests during the verification step is very costly in both computation price and time.
- Inference speed: calling multiple inferences at runtime is very slow, making use cases with real-time requirements impractical, especially if the AI API experiences networking issues or is not synchronized with Oracle. Additionally, calling multiple APIs for a single input is expensive (duplicating fee for one result).
AI Oracle 2.0
With the growing large size of AI models (e.g., common LLMs have 7 billion parameters, a larger model like Grok has 314 billion parameters (!)), and the proprietary nature of the model (i.e., totally black-box or partially open-source), doing multiple verification steps and calling to aggregate the inference results are impractical.
AI Oracle 2.0 with Distilled AI layer will provide a game-changing architecture for the previous problems:
- Aggregating AI before inference execution
- Offloading heavy work to computation-intensive layer so that Oracle mechanism can be more specialized and easier to maintain.
- Distilled AI layer is still decentralized and provides trustworthy proofs for AI.
- Enabling a paradigm-shift for AI development: Private Intelligence.
Documentations
Development and Integration documentaions are being prepared. With the separation of Distilled AI as an independent layer, we’re resolving architecture & system consistency.
Stay tuned for the update.