Over 9.6k subscribers
I hope all 7989 of you are having a great week so far 🔥
If you're enjoying my writing, please share Terminally Onchain with your friends in Crypto and join the /toc channel on Farcaster 🤝
Today's post is not a deep dive but more so a resource guide + my notes on a topic that's new to me.
Hopefully it can serve as a good starting point to learn more if you're curious.
AIs need Stablecoins
Internet of Intelligence
Resources & Questions
Last week, I saw Brian Armstrong retweet this Coinbase Developer article about AI agents successfully transacting with crypto wallets and stablecoins.
Around the same time, as I was doing research for my Stablecoin Framework post, I saw this tweet by Jermey Allaire (CEO of Circle).
To be honest, I haven't really looked into the crypto <> AI intersection at all. But I did find it interesting to think that there's an argument that AIs will be the top "target user" of stablecoins. Even when I was writing my stablecoin post, the context was solely focused on how USDC & USDT help folks with currency devaluation, global liquidity for products, etc. Not as the primary currency for AI agents to micro-transact with.
But when you take a step back and think about it, it seems easier to make the argument that the "hockey stick growth" of stablecoins might be catalyzed by the AI use case more so than humans at least in these next few years. With people - there's friction of having to convince people of utility, onboard merchants, etc. With AI, it's more of a developer solution...the blocker is it simply working efficiently. Right now, the biggest challenge with end-to-end AI agents is not having the ability to take action - pay for services, receive payments, etc.
Note: loose thoughts, need to think more deeply about the paragraph above...would be curious to hear the counters here in the /toc channel.
So, I finally decided to take some time and see what all is even going on in this vertical of...uh...decentralized AI? To get started, I messaged the TOC group chat asking if anyone had good starting points.
Sean Brennan (founder of ChainAgent) shared this great write-up by Cyber Fund. I went through the post and wanted to share the tl;dr and highlights with you. Then, in the last section, I put together a bunch of links from my bookmarks in case you're interested in diving deeper.
Let's get into it.
It seems like the early days of decentralized AI have been centered around token gated chat bots, GPU rental services, & LLM based autonomous agents. However, this is just the beginning.
The Cyber Fund team believes the end game is a flushed out Internet of Intelligence (IOI).
"Internet of Intelligence is an environment where software and hardware gain agency. Unlike ML models or hardware, IOI can actively accomplish tasks for you. It seamlessly routes between agents and services, employing the most relevant and capable AI systems to deliver results. Indistinguishable from magic, yet already possible technologically."
In order for the Internet of Intelligence thesis to play out, it's essential that there's a way to conduct financial transactions. Or else, there's no way to exchange value and complete tasks! We can't have E2E agents without giving them the ability to take charge and make decisions, even those that involve set allowances/budgets. If you're the CEO of a company, do you grow by delegating or micromanaging every transaction & decision?
After going through the post and having a few discussions, here's my high level understanding of the different components coming together in this decentralized AI vertical.
On the crypto side:
Crypto Wallets - the bank account. Even a human executive assistants need corporate cards right? Of course, managers set allowances, have complete visibility into transaction history, and can shut down cards at anytime. Similar kind of vibe.
Privacy Protocols - trust & safety mechanisms. Think of this more as a slider scale, less as a light switch. You don't just turn privacy on and off but rather as customized privacy settings that can serve a variety of situations.
Decentralized marketplace - the bazaar. Demand side is people like you and me wanting to get tasks done. Supply side is AI agents competing to get the tasks done efficiently for low costs.
Stablecoins - the medium of exchange. Well, AIs can't open up their wallets and hand you a five dollar bill right? What better currency to use than stablecoins: programmable money that is pegged to the world's reserve currency. See the 5 pros from the NovaNet blog post (linked in the resource section below).
On the AI side:
E2E Agents - the AI models completing the work, not helping humans with the work. These agents will come in all sorts of flavors but the underlying notion is that all parts of the task are taken care of: data input, processing, decision-making, output generation in a continuous flow. Watch this Devin video from Cognition Labs.
Some things Devin can do (from the website):
After reading a blog post, Devin learns how to run ControlNet on Modal to produce images with concealed messages
Build and deploy apps end to end
Can train and fine tune its own AI models
Autonomously find and fix bugs in codebases
Address bugs and feature requests in open source repositories
Multi-agent routers - the conductor of AI models. Just as we have an endless number of websites today, we'll have an endless number of models. Similar to DNS routing, we'll need AI routers to orchestrate the process from the moment a task request is sent in to the end output.
Decentralized compute - distributed infra layer of IOI. Compute for IOI is like what the Cloud is for the Internet --> instead of a "AWS of compute" how can we get a hardware to serve similarly as Ethereum nodes?
Open source models - community owned models with proper funding & royalty mechanisms baked in through tokens (ICO type vibes?).
Misc. thoughts & questions
What % of stablecoin usage will be AIs in the next 3 years? 5? 10?
If everyone has access to these models, E2E agents, etc. then what becomes the differentiator? How do you think about cost, margins, competition, etc.?
I liked the framing of "compute is the next cloud" as a beginner framework to understand AI infra. Need the reliability, speed, etc. of cloud services today but with the infra being distributed through decentralized networks
How easy will it be for AIs to bypass guidelines & restrictions in terms of "task allowances" and privacy settings? I guess this is where the MPC (multi-party computation) comes in?
How should we think about APIs in the Internet of Intelligence? Seems like they'll be increasingly important. What's different with how AIs vs humans interact with them?
It really hit me today - even after hearing about this since GPT first came out - that personalization of everything
Are "AI models" effectively the websites in IOI? How do we think about the aggregators? Do models become as niche as websites? Are there WordPress type tools that pop up?
Bookmarks
Follow Stephan (Cyber Fund) on Twitter
https://docs.cdp.coinbase.com/mpc-wallet/docs/ai-wallets/
https://gagra.vc/flipping-the-ai-coin
https://github.com/bigscience-workshop/petals
In the last three years, with the rise of much more powerful AI in the form of modern LLMs, and the rise of much more powerful crypto in the form of not just blockchain scaling solutions but also ZKPs, FHE, (two-party and N-party) MPC, I am starting to see this change.
de-ai
AI agents, USDC, and zero knowledge proofs are converging to bring about autonomous and trustless global payments.
The advent of public blockchains is one of the most profound advancements in computer science history. But the development of artificial intelligence will, and already is, having a profound impact on our world.
Independent "scifi" AIs need 4 things: 1) A foundation model + Agent framework for 100% success rate on complex Tool / API use: Many companies building this 2) A decentralized, uncensorable communication network: @xmtp 3) A financial system separate from human control: Ethereum + @coinbasedev MPC wallets (plus a way to guarantee they can keep a private key safe from any human, hard problem) 4) A way to guarantee their server will never go down, and can proliferate backups around the world: Akash Network and others. The below architecture uses #1 and #4, while the Chainagent SDK is focused on the new field of providing #2 and #3 for any Agent. https://warpcast.com/seanwbren/0x9c8b0320
Transformers etched into silicon. By burning the transformer architecture into our chips, we're creating the world's most powerful servers for transformer inference.
Run your own AI cluster at home with everyday devices 📱💻 🖥️⌚ - exo-explore/exo
From the TOC Chat
That's all for today's post! Hopefully this format was helpful for you.
See you all on Friday, hope everyone has a great rest of the week :)
- YB
Collect this post as an NFT.
316
Day 6: today's newsletter post wasn't really a deep dive, more so quick notes (and a bunch of links I compiled!) on a topic new to me: decentralized AI. Last week, I saw Brian Armstrong retweet a CB Developer article about AI agents successfully transacting with crypto wallets and stablecoins. To be honest, I haven't really looked into the crypto <> AI intersection at all. But I did find it interesting to think that there's an argument that AIs will be the top "target user" of stablecoins. Right now, the biggest challenge with end-to-end AI agents is not having the ability to take action - pay for services, receive payments, etc. After going through a post on the "Internet of Intelligence" by Cyber Fund that @seanwbren shared with me, here is my high level understanding of the different components coming together in the decentralized AI vertical. in the last section, I shared all my bookmarks in case you're interested in diving deeper. https://terminallyonchain.xyz/decai