H/t to ChatGPT for creating the prompt I used in Midjourney to create this AI Image.
Prompt: Craft an imaginative and compelling artwork that explores the intertwined relationship between Artificial General Intelligence (AGI) and the world of Cryptocurrency. Illustrate a futuristic scene where an advanced robot or full-blown AGI not only interacts with but integrates cryptocurrency into its existence.
Thesis
When I first tried Midjourney and then ChatGPT, their power initially terrified me. The sheer lack of understanding I had for their capabilities triggered a slight existential crisis. This led me to realize that, while LLMs only currently excel at finishing our sentences, they are inevitably one step closer to full-blown AGI that can deeply influence our thoughts.
The media’s portrayal of AGI as an all-powerful app or an iRobot often undersells its true impact. AGI will permeate every facet of your life, even without your awareness. Consider AI2041, a look to the future where a family’s AI insurance app devolves into restricting the daughter’s love interests based on caste and risk analysis of her true love. This illustrates how deeply woven and influential AI will become.
Every Sci Fi movie paints a dystopian AI future since it sells, and this is a realistic end-game. No matter how ethical or moral we assume OpenAI’s board to be, by the very nature of human existence there exists some bias in them which we can’t allow to negatively impact all of the applications and use cases built atop these foundational models. You may love talking to ChatGPT but what if you’re in court and your AI jury uses facial recognition to see the color of your skin or the unique spelling of your name to send you to prison for twice as long? The implications are unsettling.
Centralized AI is an inevitability. Once Google asks for 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 personalized AI’s per device since they are behind the global AI race and need an angle which leverages their secure brand. If OpenAI adjusts the model slightly and that ripples through society, and the millions of custom ChatGPT built atop, is that something you’re comfortable with?
We need an alternative. Crypto is the perfect marriage for AI since the transparent global human coordination that underpins the movement is something that can harness AI for good at global scale. Crowdfunding (with cash or with your GPU) the creation and fine tuning of open source models which anyone can audit in real time for biases or issues is the safest path forward in the accelerating world of AI.
I believe we are headed for a world of billions of AI models, whether it’s downloading and personalizing those open source models per person or projects and companies building their own for specific use cases (think Uniswap LP provision).
Crypto and AI is the perfect marriage with the cornerstone being the auditability, community ownership and community direction of the most powerful technology of our generation. Whether it’s using everyone’s GPU’s to train a model and giving them ownership in the model, DeFi and smart contracts leveraging AI within their use cases to expand their abilities or personalized AI tailored to you (vs one that has to generalize about the whole world like Bard) the fit is logical.
Decentralized AI will transparently share inner workings of the most powerful technology of our generation. Centralized AI just can’t offer this core value.
In the end, AGI will use Crypto since it will trust code and math vs physical bank branches and the whims of human nature. Our future evolved AI creations will use Crypto, we should too.
Crypto x AI Themes and Thoughts
Cypherpunk values for AI:
All of the core Cypherpunk values of Ethereum outlined by @VitalikButerin apply to AI: no deplatforming, open-global participation, censorship resistance, neutrality, cooperation, etc. The idea that we are rebuilding AI under the centralized guise of Web2 is laughable.
DePin and AI Is A Clear Use Case
Over the last ten years all of our research has gone into how to make hyperscale data centers more performant and efficient. Over the next ten years I expect technology to lean into latent GPU and user hardware at scale to do training and inference for AI models. There is clearly limited demand for the H100s Nvidia can ship and tech companies have a hold over what is available. Making our Mac Pro GPUs and other hardware available at scale for training and inference is an obvious use case. Leaders include @ionet_official,@akashnet_ and @gensynai. There may even be a path in the future where Nvidia turns inward and instead of selling its H100s simply builds out its own largest cluster.
Incentivizing New Model Creation
Building off app-level creativity, we need to incentivize the development of entirely new models themselves. This includes funding for training, crowdsourcing specific training data, and incentives to host models for inference. A large language model is only one type of AI model and even then there are dozens of leading ones (Bard, ChatGPT, Claude, etc). Users around the world can provide their GPUs, capital or data to train and fine tune models at scale and own part of the final model.
Better Apps with AI and Smart Contracts
Unencumbered decentralized AI will offer much better apps. Smart contracts referencing AI models can expand the design space for apps and greatly increase their logic and capabilities. Imagine Uniswap liquidity provision influenced by a massive off-chain model using ZK to ensure the model hasn’t been tampered with. Examples include @inference_labs @gizatechxyz and @ModulusLabs
Just look at @testmachine_ai which offers a predator mode to audit your crypto code in real time and learn from it vs waiting 6 months for a costly manual audit. Or just see the massive machine learning model that offers accurate pricing for NFTs via @UpshotHQ
AI Makes Crypto Easy To Use
In the future, most crypto users will never see all of the various nuts and bolts and endless acronyms and vocabulary we all talk about in the crypto sphere. They will simply type their intent into an LLM, and a network of solvers will handle all of the hard steps of their transaction. This LLM will learn, personalize and make your life easy. Few will have to ever manually bridge assets themselves, that’s for the solvers to earn a fee.
Best Model Decider Wins
I believe we are headed to a world of millions of AI models, whether it be everyone having their own personal model or every project and corporation having sets of their own. We already have 490k+ open source models on Hugging Face and 3 million custom ChatGPTs on the OpenAI App store. When a full fledged AI service in crypto serving users goes mainstream, I think the protocol that is able to effectively choose which model to use for each situation will be very valuable. If you are building this, please shoot me a message.
Just this week @NousResearch released plans for a new Bittensor Subnet that can evaluate open source models with a logical next step of this rating being used in directing requests to the right models.
Moral, Ethical and Legal Concerns Limit Centralized AI
Centralized players keep getting sued and kneecapping models due to moral and ethical concerns. Imagine a decentralized LLM paying people for their data with a sign-off, instead of the NYT suing @OpenAI. This limits centralized development compared to open systems which can just ship (i.e.@bittensor_). While centralized players wrestle over IP related lawsuits and moral and ethical concerns over shipping ever smart AGI, crypto networks can launch and deploy these networks without the bureaucracy and simply win.
Decentralized AI Offers Transparency
People will expect transparent training (“we built this model how you said”) and inference (“my request wasn’t messed with”). Centralized AI just can’t offer this core value. Even if it’s hard for everyday people to audit the model, similar to crypto, the idea is you can contract someone or an AI who can audit it.
Real-time visibility Into the future:
I think people want real-time visibility into the future of AI, not just updates when OpenAI feels like sharing. This is only possible with transparent and decentralized systems. Do you really want to figure out after the fact that AGI is discovered?
AGI and Permissionless Money:
The Utopian or Dystopian arc of AGI favors such a mind interacting with permissionless money to build out its desires. The AGI of the future won’t have a chase checking account. It will further build out decentralized AI and crypto, outside the control of the Fed or an OpenAI board.
Crypto x AI Visualization
We need more platforms offering visualization of what is going on under the hood of AI models which crypto projects are leveraging. When you use Bittensor’s Subnet 1 for text generation, how are you 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.
Tokens Incentives In An AI World
The use of tokens to drive ownership and coordination around AI projects will be interesting. Currently tokens attract supply side users (and speculators) but developers using centralized companies like OpenAI have a lock on the demand side with access to tons of users. It will be interesting to see if crypto projects can effectively bootstrap the demand side beyond the supply side token incentives.
Attracting Real AI Talent
Crypto x AI projects will have to attract real AI talent from Web2. This is a hurdle given Crypto is so-so in the minds of Web2 builders. The projects that are able to attract real AI talent from Web2 will be at a significant advantage. I just think it’s easier to learn crypto than it is to learn how to build a foundational AI model.
Delphi Ventures Crypto x AI Portfolio Companies
Delphi has been very active at the juncture of Crypto x AI. We are honored to back leading projects in the space.
-
@ionet_official: GPU clustering at scale
-
@inference_labs: Allowing DeFi and smart contacts to leverage off-chain models trustlessly
-
@0G_Labs @mheinrich: Data availability for On-Chain AI.
-
@UpshotHQ: An AI network for the next generation of decentralized applications.
-
@testmachine_ai: AI powered proprietary algorithm for auditing smart contracts
-
@taofuxyz: Liquid staking token for Bittensor
-
@MythosVentures: Leading early stage AI VC fund
-
@altstatemachine: Unique Metaverse AI you can own, train and trade
-
@GeppettoAi: AI game and video creation
-
@StabilityAI: Open Source tools for AI, through Seed Club VC
-
@mypeachai: NSFW companion (Angel)
If you’re building at the juncture of Crypto x AI, shoot me or anyone at Delphi Ventures a DM.
For more resources, check out our recent podcasts with UpShot and Bittensor.