Original title: ORA: Ethereum’s Trustless AI
Author: ORA
Compiled by: Fairy, ChainCatcher
Editor's note: On June 26, ORA, an oracle protocol, announced that it had completed a $20 million financing round, with investors including Polychain, SevenX Ventures, HF0, and Hashkey Capital. At the intersection of blockchain and artificial intelligence, ORA is opening up new frontiers. By combining AI, rich data sources, and computing power at any scale, ORA not only breaks the limitations of smart contracts, but also gives developers room for innovation.
This article details ORA's core technologies and products, including zero-knowledge machine learning (zkML), optimistic machine learning (opML), and optimistic privacy-preserving artificial intelligence (opp/ai). Read this article to fully understand how ORA promotes the development of on-chain AI.
Introduction to ORA
ORA is a trustless AI for Ethereum that enables AI and computation at any scale on the blockchain.
ORA breaks the limitations of smart contracts by providing AI, richer data sources, and arbitrary computing so that developers can innovate freely.
ORA's work is trusted by Compound, Ethereum Foundation, Uniswap, Optimism and others.
1. Oracle ORA on-chain artificial intelligence
ORA makes on-chain artificial intelligence a reality. ORA has contributed a series of cutting-edge research and products to the field of on-chain artificial intelligence. Products include zkML, opML and opp/ai.
a) zkML: Zero-Knowledge Machine Learning
zkML is a cryptographic approach to on-chain machine learning.
ORA is a pioneer in the field of zkML and invented the world's first zkML library: Keras2Circom.
Keras2Circom outperformed other zkML frameworks according to a research benchmark supported by the Ethereum Foundation.
Furthermore, Keras2Circom is production-ready and battle-tested in a rich ecosystem.
Read more here and check out the Keras2Circom repository on GitHub.
b) opML: Optimistic Machine Learning
opML is a cost-effective machine learning framework that now enables on-chain machine learning.
ORA is the inventor and creator of opML, both from an academic research and open source implementation perspective.
zkML can generate cryptographic proofs for ML computations that are compact enough to be verified on-chain. However, current computing power is not practical to generate proofs efficiently and economically, especially for large AI models.
In contrast to zkML, opML can efficiently bring large models like Grok’s 314 billion parameter model on-chain, reducing management costs by more than 1,000,000x.
Learn more in our paper: https://arxiv.org/abs/2401.17555.
c) opp/ai: Optimistic privacy-preserving artificial intelligence
opp/ai, optimistic privacy-preserving artificial intelligence, combines zkML for privacy protection with opML for efficiency.
opp/ai is a hybrid model tailored for on-chain AI.
ORA is the inventor of opp/ai.
Compared to pure zkML, opp/ai performs better while maintaining the same privacy properties as zkML.
Read more here and learn more in our paper: https://arxiv.org/abs/2402.15006.
2. ORA On-chain Artificial Intelligence Oracle (OAO)
a) Introduction
ORA’s Onchain AI Oracle (OAO) is ORA’s verifiable, decentralized AI oracle.
OAO enables anyone to use on-chain AI reasoning on any blockchain.
OAO uses opML to create verifiable proofs for ML computations that can be verified on the blockchain. This opens up new possibilities for AI in the blockchain.
b) OAO on Ethereum and Optimism
OAO has been running on Ethereum and Optimism based on opML to generate verifiable AI reasoning.
Last month, ORA OAO was deployed on the Ethereum mainnet.
Recently, in partnership with Optimism, ORA OAO has been integrated into the Optimism mainnet.
Read more about the OAO deployment announcement on Ethereum and Optimism.
c) Build using OAO
As a developer, you can check out the following resources to build AI-powered on-chain applications:
OAO Buy now: OAO.
OAO Documentation: AI Oracle.
Video tutorial on building with OAO: How to interact and build with OAO.
Ideas built with OAO for inspiration: So you want to build with Onchain AI.
ORA ecosystem project reference: awesome-ora.
3. ORA Initial Model Release (IMO)
In the AI era, ORA introduced a new mechanism, IMO (Initial Model Offering).
a) Introduction
TL;DR: IMO tokenizes AI models on-chain.
For AI models, IMO provides sustainable funding for open source AI models.
For the ecosystem, IMO helps align values and incentives for distribution and continued contribution.
For token holders, IMO allows anyone to capture the value of on-chain AI models from sources such as on-chain revenue and inference assets (e.g., ERC-7007).
Many open source AI models face challenges in monetizing their contributions, resulting in a lack of motivation for both contributors and organizations. As a result, the AI industry is currently dominated by closed-source, for-profit companies. The winning formula for open source AI models is to raise more funding and build them publicly.
With IMO, we can win the battle for open source AI. IMO can enable sustainable development of the open source AI model ecosystem by promoting long-term benefits, encouraging participation and funding the open source AI community. When we have an open source model that is better than a proprietary model, we win.
IMO tokenizes ownership of open source AI models and shares their profits with token holders.
b) Core components of IMO
IMO is permissionless, so anyone and any community can IMO their AI models.
IMO, there are two core components:
On-chain AI model with verifiability
On-chain revenue sharing
For Onchain AI, we need to tie the token to the right AI model and inference. ORA is doing this through the aforementioned ORA OAO, the world’s first AI Oracle that works with any AI model.
For on-chain usage revenue sharing, ORA introduced the ERC-7641: Intrinsic RevShare token as the basis for IMO to achieve fair revenue sharing. IMO tokens are not AI meme coins, but represent shares of AI models. Holders can benefit from the revenue of AI models from OAO and reasoning assets such as ERC-7007: Verifiable AI Generated Content Tokens.
c) IMO Overview
IMO is permissionless, so anyone and any community can IMO their AI models.
IMO tokenizes specific AI models, providing:
Communities are able to effectively raise funds for open source.
Contributors are incentivized to continue improving the model for global accessibility.
Token holders can gain income opportunities by using the on-chain model.
A meaningful IMO requires a revenue sharing mechanism for Onchain AI and IMO tokens:
Onchain AI can be realized through ORA OAO.
Revenue is derived from the use of models in ORA OAO, and from inference assets such as ERC-7007 co-authored by ORA.
The revenue sharing mechanism can be implemented through ERC-7641 written by ORA.
Read more about IMO here and here .
Watch ORA’s announcement of IMO at ETHDenver.
Watch IMO explained in 60 seconds.
4. OpenLM IMO: The world’s first IMO
$OLM on the mainnet: 0xe5018913F2fdf33971864804dDB5fcA25C539032LP Address on the mainnet: 0xB57CFe32dbF5227d9c9A03eF3D2E9E53cb4524B0
a) Introduction to OpenLM
OpenLM is an executive language modeling (LM) repository that aims to facilitate research in medium-sized LMs.
OpenLM is open, high-performance, decentralized AI, not the closed, centralized, permissioned LLM funded and overseen by big tech oligarchs.
OpenLM is an open source model under the MIT license created by a decentralized team. Their contributions will be recognized through this donation, which will also help to incentivize and promote the development of open source LM that anyone can access.
b) OpenLM IMO
On April 10, we released the world's first IMO: OpenLM IMO.
OpenLM IMO tokenizes OpenLM's open source model.
Just as Prometheus stole fire for civilization, ORA demonstrated the benefits of IMO for the open source AI community through OpenLM’s first IMO
$OLM is not a token of ORA. It is a tokenization of the OpenLM model on Ethereum, and ownership is shared by OpenLM contributors and the community. Nevertheless, ORA's on-chain AI oracle can enable the tokenization of artificial intelligence models.
IMO will be a cornerstone of the OpenLM journey, providing funding and incentives to foster a sustainable ecosystem, democratize access, and drive continued development.
IMO for OpenLM promises to revolutionize the LLM landscape by leveraging the power of open source software and the community.
Read more about OpenLM IMO here.
c) OpenLM IMO activities
The first OpenLM IMO worldwide was a huge success.
All 500 million tokens were sold for 150 ETH (~$500,000). This happened in 5 minutes. This proves that demand is high given the low supply. $OLM continues to change hands on the decentralized exchange Uniswap.
Unfortunately, the vast majority of people who attempted to purchase $OLM when the sale began had their transactions not successfully processed, as demand far outstripped the limited supply.
This is the first IMO, there will be more to come. Future IMOs will have different parameters to allow more valuable people and wallets to participate.
d) OpenLM IMO Roadmap
The goal of IMO is to foster the open source community and grow the ecosystem. OpenLM IMO provides more resources for the OpenLM open source community.
This is a flagship demonstration of IMO technology. This is the beginning of the journey towards a more sustainable, competitive, and open source AI future.
Here is the roadmap to achieve OpenLM IMO success.
1. OAO Integration
In order to achieve revenue sharing of $OLM, ORA integrated the OpenLM AI model into the Onchain AI Oracle. In this way, every time a user interacts with the OpenLM model, on-chain revenue can be generated and tracked by OAO.
2. OpenLM ecosystem expansion
For AI models, IMO provides sustainable funding support for open source AI models. With this effort of OpenLM IMO, various projects can use OpenLM to contribute to the $OLM revenue pool.
3. Donate to the OpenLM community
As previously announced and planned, 10% of $OLM IMO tokens will go to existing and future OpenLM contributors.
We are finalizing all the details of our support for an open and decentralized AI community.
4. Revenue Sharing Activities
The OpenLM RevShare token ($OLM) has the following features:
Revenue Sharing: Part of the revenue from OAO or other sources will be allocated to the revenue sharing pool of $OLM tokens. Token holders can claim the revenue every 90 days after the snapshot.
Burn: A portion of revenue from OAO or other sources will be allocated to the $OLM token burn pool. Token holders can burn their tokens as a buyback mechanism. The burn mechanism will support the "floor price" of OpenLM tokens.
