Delphi Digital: Why the combination of encryption and AI is inevitable

Author: Delphi Digital Lian Tommy , Source: Author Twitter @Shaughnessy119; Compiler: Deng Tong, Golden Finance

When I first tried Midjourney and later ChatGPT, their powerful capabilities were initially It scares me. My total lack of understanding of their abilities triggered a minor existential crisis. This made me realize that while current LLMs are only good at refining our sentences, they are inevitably closer to a fully developed AGI that can profoundly influence our thinking.

Media portrayals of AGI often underestimate its true impact, treating it as a do-it-all application or iRobot. AGI will permeate every aspect of your life without you even realizing it. Imagine AI2041, a vision of the future where a family’s AI insurance app evolves to limit a daughter’s romantic choices based on social class and a risk analysis of true love. This illustrates how profound and impactful AI will become.

Every science fiction movie depicts a pessimistic artificial intelligence future because such plots are more popular and have a realistic ending. No matter how moral and ethical we think the OpenAI board is, due to the very nature of human existence, there are biases within it that cannot allow it to negatively impact all applications and use cases built on top of these foundational models. You may enjoy communicating with ChatGPT, but how would you feel if you were in court and your AI jury used facial recognition to see your skin color or the unique spelling of your name to send you to jail for double the time? These effects are disturbing.

Centralized artificial intelligence is inevitable. Once Google asks your permission to access your gDrive, gDocs and Gmail, your personalized AI will take on a life of its own. I expect Apple to launch one personalized AI per device because they are lagging behind in the global AI race and need a product that leverages their security brand. Would you feel comfortable if OpenAI fine-tuned the model and this had a ripple effect on society, affecting the millions of custom ChatGPTs built on top of it?

We need an alternative. Cryptocurrency is a perfect fit for AI because the transparent global human coordination that underpins this movement can leverage AI to good effect on a global scale. Crowdfunding (with cash or GPUs) to create and fine-tune open source models that anyone can audit for bias or issues in real time is the safest path forward in the accelerating world of AI.

I believe we are heading towards a world with billions of AI models, whether that's open source models downloaded and personalized for everyone, or projects and companies building their own collections of models for specific use cases (think Provided by Uniswap LP, Exchange Risk Analysis, Delphi Artificial Intelligence) Analyst).

Crypto and artificial intelligence are a perfect match, the cornerstones of which are auditability, community ownership, and community orientation of the most powerful technology of our generation. Whether it’s leveraging everyone’s GPUs to train models and giving them ownership in the models, or DeFi and smart contracts leveraging AI in their use cases to expand their capabilities, or personalized AI that’s customized for you instead of For an AI like Bard that must generalize about the entire world), this combination makes sense.

Decentralized AI will transparently share the inner workings and ownership of the most powerful technology of our generation. Centralized artificial intelligence cannot provide this core value.

Ultimately, AGI will use cryptocurrencies because it will trust code and mathematics rather than physical bank branches and the whims of human nature. Our future evolving AI creations will use cryptocurrencies, and so should we.

Cryptox AI themes and ideas

Cryptopunk’s artificial intelligence values:

Cypherpunk values ​​at the heart of Ethereum, as outlined by @VitalikButerin, apply to AI: no de-platforming, open global participation, censorship resistance, neutrality, collaboration, etc. The idea that we are rebuilding artificial intelligence under the centralized guise of Web2 is laughable.

AGI and NoneLicensed currencies

AGI’s utopia or dystopia favors an idea that interacts with unlicensed money to realize its ideas. Future AGIs will not have Chase checking accounts. It will use cryptography to further build decentralized artificial intelligence and cryptocurrencies without the control of the Federal Reserve or OpenAI’s board of directors.

DePin and AI are a clear use case

In the past decade, all of our research All focus on how to improve the performance and efficiency of hyperscale data centers. Over the next decade, I expect technology to leverage underlying GPU and user hardware at scale to train and infer AI models. There's obviously limited demand for the H100 that Nvidia can ship, and the tech company has control over what's available. Making our Mac Pro GPUs and other hardware available for training and inference at scale is an obvious use case. Leaders include @ionet_official, @akashnet_ and @gensynai. There might even be a path forward where Nvidia turns in-house and instead of selling its H100, simply builds its own largest cluster.

Incentivize the creation of new models

To build application-level creativity, we need to incentivize the development of entirely new models themselves. This includes training funding, crowdsourcing specific training data, and incentives for hosting inference models. Large language models are just one type of AI model, and even then there are dozens of leading models (Bard, ChatGPT, Claude, etc.). Users around the world can provide their GPUs, funding, or data to train and fine-tune models at scale and own a piece of the final model.

Smarter applications and smart contracts with artificial intelligence

Unencumbered decentralized artificial intelligence Intelligence will provide better applications. Smart contracts that reference artificial intelligence models can expand the design space of an application and greatly enhance its logic and functionality. Imagine that Uniswap liquidity supply is affected by a large-scale off-chain model using ZK to ensure that the model cannot be tampered with. Examples include @inference_labs @gizatechxyz and @ModulusLabs , just look at @testmachine_ai which provides a predator mode to audit your crypto code in real time and learn from it instead of waiting 6 months for an expensive manual audit. Or check out the large-scale machine learning models that provide accurate pricing for NFTs via @UpshotHQ.

Artificial intelligence makes cryptocurrency easy to use

In the future, most cryptocurrency users will never You won’t see all the minutiae, endless acronyms and vocabulary we talk about in the cryptocurrency space. They simply feed their intent into a low-level machine learning (LLM) and a network of solvers will handle all the complex steps of their transaction. This LLM will learn, personalize and simplify your life. There is rarely a need to manually bridge assets, it is the solver that earns the fee.

Best model decision maker wins

I believe we are heading towards a world with millions of A.I. In the world of models, whether everyone has their own personal model or every project and company has its own model. We already have over 490,000 open source models on Hugging Face and 3 million custom ChatGPTs on the OpenAI app store. When mature AI services become mainstream among crypto service users, I think protocols that can effectively choose which model to use for each situation will be very valuable.  

Just today, @NousResearch announced plans for a new Bittensor subnet that can evaluate open source models and use the logical next step of that rating to direct requests to the correct model.

Moral, ethical, and legal issues limit centralized artificial intelligence

Centralized actors are constantly faced with litigation over moral and ethical issues, and thus undermine the model. Imagine a decentralized low-level machine learning (LLM) that pays people for their data by signing agreements, rather than like the New York Times filing a lawsuit against @OpenAI. This limits the development of centralization compared to centralized systems, where open systems can be released directly (e.g. @bittensor_). At a time when centralized players struggle with intellectual property-related litigation and moral and ethical concerns about releasing smarter artificial intelligence, cryptonetworks can launch and deploy these networks without red tape and win with ease.

Decentralized AI provides transparency

People will expect transparent training (“We do what you want saying the model was constructed") and reasoning ("my request was not messed up"). Centralized AI cannot provide this core value. Even though it would be difficult for normal people to audit the model, similar to cryptocurrencies, the idea is that you can sign a contract with someone or an AI who can audit the model.

Real-time insights into the future

I think people want to understand the future of artificial intelligence in real-time, not just in OpenAI updates when it wants to share. This is only possible through a transparent and decentralized system. Did you really want to find out after AGI was discovered?

Crypto x AI Visualization

We need more platforms to deliver the artificial intelligence that crypto projects are leveraging A visualization of what is happening behind the model. When you use Bittensor's subnet 1 for text generation, how do you make sure it's not just running your prompt through Bard or ChatGPT? I'm not saying this is a bad thing, but I don't know the answer.

Token incentives in the AI ​​world

Use tokens to drive ownership and ownership of AI projects It will be fun to coordinate. Currently, tokens attract supply-side users (and speculators), but developers using centralized companies like OpenAI lock in the demand side and can reach a large number of users. It will be interesting to see if crypto projects can effectively steer the demand side beyond supply side token incentives.

Attract real AI talent

The Crypto x AI project must attract real AI talent from Web2 . This is a hurdle given the mediocre performance of cryptocurrencies in the minds of Web2 builders. Projects that can attract real AI talent from Web2 will have a significant advantage. I just think it's easier to learn about cryptocurrencies than it is to learn how to build basic AI models.

Delphi Ventures Crypto x AI Portfolio Company

Delphi excels at the intersection of Crypto x AI Very active. We are proud to support leading projects in this field.

- @ionet_official: GPU clustering for large-scale heterogeneous hardware.

- @inference_labs: Allows DeFi and smart contracts to leverage off-chain models through ZK.

- @0G_Labs @mheinrich: Data availability for on-chain artificial intelligence.

- @UpshotHQ: An artificial intelligence network for the next generation of decentralized applications.

- @testmachine_ai: AI-powered proprietary algorithm for auditing smart contracts

- @taofuxyz: Liquid staking tokens for Bittensor and more.

- @altstatemachine: Unique Metaverse AI you can own, train and trade.

- @GeppettoAi: Artificial intelligence game and video creation.

- @StabilityAI: Open source tools for artificial intelligence, via @seedclubvc (h/t Nima and Anthony).

- @MythosVentures: Leading early-stage AI venture capital fund.

- @mypeachai: NSFW companion (angel).