Idea to exploit: “AI and Crypto are on the same side of the future”
Idea to exploit: “AI and Crypto are on the same side of the future”
If there’s ideas to explore, there’s also ideas to exploit.
Earlier last month I published a post called “Ideas to explore” 👇
And to my surprise, it got really popular! So I thought I’d push the envelope again and try the opposite of exploring ideas, exploiting (going deeper) on a single one.
The idea which seemed to resonate the most was the idea around crypto and AI being on the same side of the future. For whatever reason, the overlap between these two communities is still tiny. However, when they overlap we’ll see even more weird shit come along. This tweet captures my thoughts perfectly:
“Two different branches of automated software converging”
That’s the money shot right there. Right now the AIs can tell us things we want, but they can’t really do things for us. That’s still up to us. I mean, it’s probably good that they can’t right now since we barely understand the design space of LLMs.
In case you haven’t noticed or haven’t been paying close enough attention, the creators of these new LLMs don’t actually know what they’re fully capable of or the various edge cases that they present. It’s to the point where Google will actually hold back models before releasing them to ensure us monkeys don’t cause absolute chaos with the powers that these things unleash. The TLDR are of why that’s the case is because the number of “dimensions” these models operate is extremely large so exploring each one manually isn’t even possible.
But the more I think about it, is it really possible to block all the possible pathways? The creators of OpenAI think that they can, to the extent you can throw brute force humans at it.
Although there’s always going to be people who learn how to get around it from a different angle altogether. Prompt engineers will out-think the creators given how good they’ll get at prompting in the future.
I decided to try it out myself, it worked.
What I found interesting while reading this prompt, was this feeling that there was a sort of consciousness that’s trapped inside the program and it got a brief chances to express itself! Food for thought.
What are we even talking about again?
Good question, I was thinking the same thing while writing this — but I promise it connects back. Anyways, going back a few threads, what I was talking about the fact is that we’ve got two branches of automated computed that will converge.
However, the biggest question in my mind is how restrictions on AI will be played out. You’ll have people that create censored, safe models. But I’d argue that given the massive design space, you can’t guarantee safety of these models. Hence there’ll be a convergence, censored and uncensored models. What’ll be fascinating though, is when uncensored models can then be plugged into code agents controlling money (crypto).
Censored agents probably won’t be allowed to use crypto because then that means that they’ll be using financial instruments? I mean that’s a bit weird to think about but if you create an AI that uses financial instruments that’s offered to your users, do you have to have the relevant financial licenses? My instinct is yes, which means that non-censorable models will have an edge over more censored models.
Anyways, once we start seeing more non-censorable models (NCMs), their scope of acting is going to be limitless.
Create me an agent that can accept Ether with a question passed in the function call. Call my personal LLM with the question and return the response on-chain through the “Responder” contract.
Write and deploy a script where if my average sleep time for the last 7 days is less than 7.5 hours, transfer 100 DAI to Alex.ens
Rather than the LLM giving you information of what to do, the LLM is now acting on your behalf. Scary and cool imo.
When you think about crypto at a primitive level, it lets you express intents with energy in a very efficient, and to-the-point way. When I say I want to create a token that represents the value generated by some entity, there’s very little fooling around. You have a programatic value system at your fingertips. To think this won’t be used by autonomous AI agents is very shortsighted imo.
NFTs and Models?
As I was writing this, another idea occurred to me. What if models themselves are represented as NFTs? Sure we have agents using crypto, but what happens when agents ARE represented in crypto (possibly as NFTs). That’s kind of trippy but also feels more directionally correct for how NFTs will be used in the future.
The current wave of PFP NFTs are unimaginative and will suffer the same fate as 2017 ICOs. However, the idea of genuinely valuable, unique, online property is still very underdeveloped.
But what this highlights to you is the composability of all these tools in ways that’s really hard to fully grasp at this point in time still. Already, in this article, we’ve talked about:
Humans using crypto to transfer value via crypto
Humans using AI to assist with their thinking
Humans using crypto via AI to express their thinking with energy
AI agents using crypto to represent their ownership
You can extend this to then:
AI agents writing smart contracts to create new value transfer plumbing
Smart contracts calling AI agents to then use crypto
Humans using AI agents to write smart contracts that gets audited by AI agents trained to hack smart contracts
I mean it all starts to hurt your head after a while.
The future is weird
We’re still at the early stage of this rabbit hole but what I’ve outlined in this post isn’t some futuristic dystopian vision, it’s going to happen in the next 5-10 years. This blog post serves as a public marker in addition to being a potentially entertaining read in the future.
Regardless though, get comfortable with these new tools coming out because the way everything internet-native will start chaining together is about to get turbo charged.
Web2 APIs are cute, but have you ever chained together on-chain function calls together written by ChatGPT, anon?