AI-powered Ethereum: from rules-based smart contracts to AI-powered smart contracts.
Ethereum just got smarter.
Eternal AI brings onchain LLM to builders so they can write contracts & launch dapps that don’t just run but learn and think.
Whitepaper: eternalai.org/ethereum.pdf
1. Truly smart contracts
Today, dapps are rule-based programs coded as smart contracts. They're like rigid machines, unable to adapt or learn.
To address this, we must rethink our approach to dapps and create a framework for devs to add AI capabilities to their contracts.
2. Rule-based programming model
Consider a contract that manages a decentralized fantasy sports league. It needs to determine the winner between the two teams.
With rule-based, the contract uses a set of if-else statements to analyze the performance & determine the winner.
3. AI-powered programming model
In contrast, the AI-powered approach enables a new programming model that uses large language models (LLMs) to make decisions dynamically in real time.
The fantasy sports league contract could be written to provide a simple prompt to the LLM.
4. AI Kernel Architecture
To build truly smart contracts & AI-powered dapps, we need a decentralized protocol that facilitates the integration of AI inference, AI models & GPU resources.
This is where the AI Kernel comes in – the central component of the AI-powered Ethereum.
5. User Space
Consider the user's journey to interacting with the AI Kernel. It starts with a prompt — a request for the AI Kernel to generate an output.
Along with the prompt, the prompt context is constructed based on onchain data and also sent to the AI Kernel.
6. Kernel Space
The AI Kernel is the heart of our decentralized AI architecture.
The AI Kernel consists of several smart contracts that work together to manage resources and facilitate communication between different parts of the protocol.
7. Model Space
The Model Space comprises two key components: AI Models and AI Model Drivers.
AI Models are well-known open-source models like Llama, FLUX, and Hermes.
AI Model Drivers adapt these models to the blockchain environment.
8. Hardware Space
The Hardware Space is where the actual computation happens in the AI Kernel.
It consists of GPU nodes that serve as the atomic compute unit of the system.
These nodes are responsible for taking user prompts, running inference, and returning outputs.
9. Proof-of-Compute
The protocol runs on a novel consensus algorithm called Proof-of-Compute.
Instead of solely solving a complex mathematical puzzle like PoW, GPU nodes perform meaningful computations on prompts requested by real users and real dapps.
10. AI-Powered Decentralized Applications
As we integrate AI into Ethereum, a new paradigm for dapps begins to take shape.
No longer limited by rigid, rule-based programming constraints, developers can create dapps capable of adapting, learning, and evolving over time.
11. Onchain Conversational AI Agents
With just a few lines of code, you can build an autonomous agent on top of the AI Kernel and earn passive income by charging a service fee when someone uses your agent.
12. AI-Powered Crypto Wallet
In this example, we’re building an AI-powered wallet.
By providing the AI Kernel with a rich context of transaction history, the model can learn to identify potential red flags, such as unusual transactions, inconsistent spending behavior & more.
13. AI-Powered Oracles
We're building an AI-powered oracle for BTC prices.
By providing the AI Kernel with a rich context of BTC price feeds added continuously by Oracle feeders, the AI Kernel can learn to return the current BTC price.
14. AI-Powered DAOs
By feeding the AI Kernel a continuously updated context of proposal result history, the AI Kernel can develop an understanding of successful and unsuccessful proposals.
This enables it to assess and predict the viability of new proposals.
15. AI-Powered Wallet Credit Scores
By providing the AI kernel with a comprehensive context of a given address's transaction history, the model can learn to accurately assess the address's creditworthiness and provide a corresponding credit score.
16. AI-Powered ENS
In this example, we're building an AI-powered ENS generator.
The model can generate an available ENS domain that best fits the given description.
If a suggested domain has already been taken, it will continue to retry until it finds a suitable one.
17. Conclusion
The integration of AI and crypto represents a paradigm shift in our approach to dapps.
Our work provides a framework for running AI onchain, opening up new possibilities for AI-powered dapps & their potential to reshape our world.
Paper: eternalai.org/ethereum.pdf
PS: Implementation
We've chosen @base as the first EVM chain to implement this protocol.
Base is ideal for this experiment because of its low cost & fast block time. The team also loves Crypto x AI.
Interact with AI-powered smart contracts on @base at .eternalai.org
Share this Scrolly Tale with your friends.
A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.