In Vitalik's latest article, he discusses the crossover between AI and Crypto, outlining two main objections: cryptographic overhead and black-box adversarial machine learning attacks.

Vitalik sees promise in the AI x Crypto direction. AI can play a key role, such as being an interface to the game or defining the rules of the game, in helping crypto become better.

While Vitalik sees great promise in the synergy between AI and Crypto, he highlights one major objection: cryptographic overhead. The most mainstream on-chain AI/ML approach today is zkML, which compiles ML models into zk circuits, allowing the cryptographic proof of computation to be verified onchain.

"AI computation is expensive already," and adding cryptography to it further slows it down.

Vitalik believes that the issue of cryptographic overhead has been partially addressed:

• AI computation and its cryptographic overhead can be significantly accelerated, and it does not involve "unstructured" types of computation like zkEVM.

• Over time, more efficient zk cryptography schemes will be invented, reducing the overhead significantly.

However, this approach is nowhere close to being practical, especially for the use cases described by Vitalik. Here are some examples:

• The zkML framework EZKL takes around 80 minutes to generate a proof of a 1M-nanoGPT model.

• According to Modulus Labs, zkML has >>1000 times overhead than pure computation, with the latest report being 1000 times.

• According to EZKL’s benchmark, the average proving time of RISC Zero is of 173 seconds for Random Forest Classification.

In practice, having to wait a few minutes for a human-friendly explanation of your tx is unacceptable.

At the end of the article, Vitalik mentioned, "I look forward to seeing more attempts at constructive use cases of AI in all of these areas, so we can see which of them are truly viable at scale." We believe that zkML is not "viable" at this stage to realize the above applications.

As the inventor of opML and the creator of the first open-source implementation of opML, we believe that opML can make AI x Crypto happen right now by completely solving the cryptographic overhead through game theory.

AnyTrust assumes that for every claim, there is at least one honest node, ensuring that either the submitter or at least one verifier is honest. Under AnyTrust, safety and liveness are preserved:

• Safety: A single honest validator can enforce correct behavior by challenging a malicious node's incorrect result, leading to penalization through an arbitration process.

• Liveness: Proposed results are either accepted or rejected within a maximum period.

When comparing "AnyTrust" with "Majority Trust," opML's "AnyTrust" model emerges as more secure. "AnyTrust" maintains a high level of security, outperforming "Majority Trust" under various conditions.

Vitalik also discussed the issue of model privacy in his article. In fact, for most models, especially the small models currently supported by zkML in practice, it is possible to reconstruct the model with sufficient inference.

Concerning general privacy, especially user privacy, opML may appear to lack inherent privacy features due to the necessity of keeping challenges public. However, by combining zkML and opML, we can achieve the optimal level of privacy to ensure secure and irreversible obfuscation.

opML can already run Stable Diffusion and LLaMA 2 directly on Ethereum. And the four categories mentioned by Vitalik (AI as player / interface / rules / objective) is already achievable with no overhead via opML.

We are actively exploring the following use cases and directions:

• AIGC NFT (ERC-7007), 7007 Studio won the Story Protocol Hackathon

• Onchain AI Games (e.g. Dungeon and Dragons)

• Prediction Market with ML

• Content Authenticity (deepfake verifier)

• Compliant programmable privacy

• Prompt Marketplace

• Reputation/Credit Scoring

With opML, we blaze new trails in AI x Crypto and remove the challenges of cryptographic overhead while preserving decentralization and verifiability.

HyperOracle is a programmable zkOracle protocol that powers smart contracts with arbitrary compute and richer data sources. HyperOracle offers full security and decentralization for trustless automation and onchain AI/ML so builders can easily create next-gen dApps.

hyperoracle.io | x.com/hyperoracle | github.com/hyperoracle

In Vitalik's latest article, he discusses the crossover between AI and Crypto, outlining two main objections: cryptographic overhead and black-box adversarial machine learning attacks.

Vitalik sees promise in the AI x Crypto direction. AI can play a key role, such as being an interface to the game or defining the rules of the game, in helping crypto become better.

While Vitalik sees great promise in the synergy between AI and Crypto, he highlights one major objection: cryptographic overhead. The most mainstream on-chain AI/ML approach today is zkML, which compiles ML models into zk circuits, allowing the cryptographic proof of computation to be verified onchain.

"AI computation is expensive already," and adding cryptography to it further slows it down.

Vitalik believes that the issue of cryptographic overhead has been partially addressed:

• AI computation and its cryptographic overhead can be significantly accelerated, and it does not involve "unstructured" types of computation like zkEVM.

• Over time, more efficient zk cryptography schemes will be invented, reducing the overhead significantly.

However, this approach is nowhere close to being practical, especially for the use cases described by Vitalik. Here are some examples:

• The zkML framework EZKL takes around 80 minutes to generate a proof of a 1M-nanoGPT model.

• According to Modulus Labs, zkML has >>1000 times overhead than pure computation, with the latest report being 1000 times.

• According to EZKL’s benchmark, the average proving time of RISC Zero is of 173 seconds for Random Forest Classification.

In practice, having to wait a few minutes for a human-friendly explanation of your tx is unacceptable.

At the end of the article, Vitalik mentioned, "I look forward to seeing more attempts at constructive use cases of AI in all of these areas, so we can see which of them are truly viable at scale." We believe that zkML is not "viable" at this stage to realize the above applications.

As the inventor of opML and the creator of the first open-source implementation of opML, we believe that opML can make AI x Crypto happen right now by completely solving the cryptographic overhead through game theory.

AnyTrust assumes that for every claim, there is at least one honest node, ensuring that either the submitter or at least one verifier is honest. Under AnyTrust, safety and liveness are preserved:

• Safety: A single honest validator can enforce correct behavior by challenging a malicious node's incorrect result, leading to penalization through an arbitration process.

• Liveness: Proposed results are either accepted or rejected within a maximum period.

When comparing "AnyTrust" with "Majority Trust," opML's "AnyTrust" model emerges as more secure. "AnyTrust" maintains a high level of security, outperforming "Majority Trust" under various conditions.

Vitalik also discussed the issue of model privacy in his article. In fact, for most models, especially the small models currently supported by zkML in practice, it is possible to reconstruct the model with sufficient inference.

Concerning general privacy, especially user privacy, opML may appear to lack inherent privacy features due to the necessity of keeping challenges public. However, by combining zkML and opML, we can achieve the optimal level of privacy to ensure secure and irreversible obfuscation.

opML can already run Stable Diffusion and LLaMA 2 directly on Ethereum. And the four categories mentioned by Vitalik (AI as player / interface / rules / objective) is already achievable with no overhead via opML.

We are actively exploring the following use cases and directions:

• AIGC NFT (ERC-7007), 7007 Studio won the Story Protocol Hackathon

• Onchain AI Games (e.g. Dungeon and Dragons)

• Prediction Market with ML

• Content Authenticity (deepfake verifier)

• Compliant programmable privacy

• Prompt Marketplace

• Reputation/Credit Scoring

With opML, we blaze new trails in AI x Crypto and remove the challenges of cryptographic overhead while preserving decentralization and verifiability.

HyperOracle is a programmable zkOracle protocol that powers smart contracts with arbitrary compute and richer data sources. HyperOracle offers full security and decentralization for trustless automation and onchain AI/ML so builders can easily create next-gen dApps.

hyperoracle.io | x.com/hyperoracle | github.com/hyperoracle