Integration of Crypto and AI: Four Key Intersection Points

Crypto and AI Integration: 4 Key Intersection Points

Author: Kyle Samani (Partner at Multicoin Capital) and ChatGPT; Translation: Blockingcryptonaitive and ChatGPT

Note: The vast majority of this article, including most of the headings, was written by ChatGPT. The author’s text is italicized. You can see the author’s conversation with ChatGPT here.

The worlds of Crypto and AI have been developing in parallel, each breaking technological and innovative boundaries in their respective fields. As we continue to make progress in both, it is becoming increasingly clear that their futures are closely intertwined. In this article, we will explore four important intersections at the crossroads of Crypto and AI.

The “AirBnB for GPUs” Model

The rise of AI and machine learning (ML) workloads has created enormous demand for high-performance GPUs like Nvidia A100. In response, a new market similar to “AirBnB for GPUs” has emerged. This allows individuals and organizations to rent out their unused GPU resources to meet the needs of AI researchers and developers.

• SharkTeam: On-Chain Data Analysis of RWA Raceway
• LD Capital: Analysis of L2 Head Liquidity Customization DEX Mechanism of Trader Joe, Izumi, and Maverick
• Aave community initiates a temperature check proposal to integrate MakerDAO’s DSR into Aave V3 Ethereum pool.

This is a truly unique moment in market history. Supply of GPUs was already unable to keep up with demand prior to the launch of ChatGPT. Since then, demand has likely grown by at least 10x, and possibly as much as 100x. Additionally, we know that models grow logarithmically with training scale; this means that demand for GPU compute grows exponentially to increase model quality. While total supply far exceeds demand, moments where demand for a good so vastly outstrips available supply are rare; if every GPU on Earth were available for AI inference and training today, there would be a surplus, not a shortage!

However, there are several key technical challenges to consider when exploring the “AirBnB for GPUs” concept:

• Not all GPUs can support all workloads : GPUs come in various shapes, sizes, and specifications. As such, certain GPUs may be unable to handle certain AI tasks. To make the model successful, there needs to be a way to match the correct GPU resources with the appropriate AI workloads. As the market matures, we should expect to see GPUs further specialize and optimize for different AI tasks.

• Adjusting the training process to accommodate higher latency : Most of today’s foundational models are trained on GPU clusters, with GPUs connected via extremely low-latency connections. In decentralized environments, latencies increase by several orders of magnitude as GPUs may be distributed across multiple locations and connected via the public internet. To overcome this challenge, there is an opportunity to develop new training processes that have higher-latency connections. By rethinking how we train AI models, we can better utilize larger, decentralized clusters of GPUs.

• Validation issues : It is impossible to know whether an untrusted computer has executed a particular piece of code. Therefore, it is difficult to trust the output of untrusted computers. However, this problem can be mitigated by combining a reputation system with cryptographic economic staking, and in some cases, by supporting novel models that allow for fast verification.

There are quite a few teams working in this field, including training and inference. Multicoin Capital has invested in Render Network, which initially focused on 3D rendering and has opened up its GPU network to also support AI inference.

In addition to Render Network, there are several other companies working in this field: Akash, BitTensor, Gensyn, Prodia, Together, and other projects still in development.

Token Incentivized RLHF (Reinforcement Learning from Human Feedback)

Token incentives are almost certainly not applicable to all use cases of reinforcement learning from human feedback (RLHF). The question is, what framework can we use to consider when token incentives make sense for RLHF and when cash payments (e.g. USDC) should be used.

Token incentives may improve RLHF as the following becomes more true:

• The model becomes more narrow and vertical (as opposed to general and horizontal, such as ChatGPT). If someone is going to provide RLHF as their primary work and therefore generate the majority of their income by providing RLHF, they may need cash to pay rent and buy food. As you move from general queries to more specific domains, model developers will need more trained staff to participate, who are more likely to achieve long-term success in overall business opportunities.

• Higher income for people who provide RLHF outside RLHF work itself. If someone has enough income or other savings to demonstrate that it is reasonable to invest time in RLHF models in a particular field, then they can only accept locked/non-liquid tokens as compensation rather than cash. To maximize the likelihood of success, model developers should not only award unlocked tokens to those who provide specific domain RLHF. Instead, tokens should be awarded over a period of time to incentivize long-term decision making.

Some industries where token incentivized RLHF models may apply include:

• Medicine: People should be able to engage in lightweight, first-response diagnostics, as well as long-term prevention and longevity medicine, with a legal master.

• Law: Business owners and individuals should be able to use large language models to more effectively navigate the complexity of various heterogeneous legal systems.

• Engineering and Construction: Enhance design tools or simulation models.

• Finance and Economics: Improve prediction models, risk assessments, and algorithmic trading systems.

• Scientific Research: Refine AI models for simulating experiments, predicting molecular interactions, and analyzing complex datasets.

• Education and Training: Contribute to AI-driven learning platforms to improve the quality and effectiveness of educational content.

• Environmental Science and Sustainability: Optimize AI models to predict environmental trends, allocate resources, and promote sustainable practices.

There is a production-ready vertical domain token incentive, RLHF: Maps. Hivemapper benefits not only drivers, but also map editors who invest time in editing and organizing map data. You can try Hivemapper’s AI training tool for maps yourself.

Zero-Knowledge Machine Learning (zkML)

The blockchain doesn’t know what’s happening in the real world. However, understanding events happening off-chain is very useful to them so they can transfer value programmatically based on the state of the real world.

Oracles solve part of this problem. However, oracles are not enough. Merely relaying real-world data to the chain is not enough. A lot of computation needs to be done before entering the chain. For example, let’s consider a yield aggregator that needs to move deposits between different pools to earn more yield. To do this in a trust-minimized way, the aggregator needs to calculate the current yield and risk for all available pools. This quickly becomes an optimization problem that is ripe for ML. However, computing ML on-chain is too expensive, so this is an opportunity for zkML.

Teams like Modulus Labs are now building in this space. We hope more teams use a general-purpose ZKVM to build in this space, such as Risc Zero and Lurk.

Authenticity in the Era of Deepfakes

As deepfakes get more sophisticated, maintaining authenticity and trust in digital media is crucial. One solution involves leveraging public-key cryptography, allowing creators to attest to the authenticity of their content by signing it with their public key.

The public key itself is not enough to solve the authenticity problem. There needs to be a public record that maps the public key to real-world identities for verification and trust-building purposes. By associating public keys with verified identities, a feedback and punishment system can be created for when someone is found to abuse their keys, such as signing deepfake images or videos.

In order for this system to be effective, integrating public key signing with real-world identity verification will be crucial. Blockchain technology, which underpins many cryptocurrency systems, can play an important role in creating decentralized and tamper-proof identity registries. The registry maps public keys to real-world identities, making it easier to establish trust and hold bad actors accountable.

There will be at least two configurations: hardware-based and user-controlled software.

• Hardware-based: We anticipate that smartphones and other devices will soon integrate hardware-based local image, video, and other media signature capabilities.

Solana Labs recently launched Saga Phone, which is powered by the Solana Mobile Stack (SMS). Over the next few months, I hope that SMS will be updated to use SMS to sign every photo using the SDK of the seed library, proving that the photo was not generated by AI.

• User-controlled software: People will use design tools such as Photoshop, Octane, and image generators such as Stable Diffusion to create artwork. We expect that these software providers will integrate public key encryption mechanisms, allowing creators to prove authenticity while also acknowledging the tools used in the creation process.

Conclusion

In summary, the fusion of cryptocurrency and artificial intelligence technology provides a wealth of opportunities to address urgent challenges and unlock innovative solutions across multiple industries. By exploring the intersections of these fields, we can find new ways to optimize resource allocation in AI training, use token incentives for specific domain reinforcement learning from human feedback, and maintain the authenticity of digital media in the face of deep fakes.

The “AirBnB for GPUs” model offers the potential for decentralized and democratized access to high-performance GPUs, enabling more people and organizations to contribute to AI research and development. Token-incentivized RLHF can be applied across industries from engineering and finance to education and environmental science to improve AI models by leveraging domain experts’ knowledge. ZKML will allow blockchains to update on-chain financial states based on complex real-world changes. Finally, by combining public key cryptography with real-world identity verification and blockchain technology, we can create a robust system to address the challenges of deep fakes and maintain trust in digital media.

As we continue to discover synergies between cryptography and artificial intelligence, we will undoubtedly find more opportunities to drive innovation, create value, and address some of the most pressing issues facing society today. Embracing the intersections of these two fields will help us break through technological barriers and shape a more interconnected, efficient, and authentic future.

We will continue to update Blocking; if you have any questions or suggestions, please contact us!