[Music] My name is Nico. I'm part of Fetch AI and Fetchi itself is part of the ASI Alliance. Four projects that share a common token together, right? Uh what we specialize in and what you're about to see is large language models working together with AI agents to execute human intents. So people have intents.
We want to build wealth. We want to start families. We want to own property. Uh we want to participate in DeFi. We want to be verified.
It doesn't really matter what the intent intent is. Some of them are high level uh and very abstract. Some of them are very specific like I want a specific car. I want to find a electronic charger or I want to use a specific D5 protocol, right? And so what we've built is a system with three components.
We're troubleshooting live as you see, but it doesn't really matter. You can just listen to my voice and that's it. Hopefully we're not going to be the only ones that have this, but it's fine. Um, the first topmost component of the system is a large language model. Think of this as a conversational UI that allows you to speak stuff into existence.
By show of hands, who is familiar with chat GPT? Really quick, everyone. Fine. So, when you prompt chat GPT, perfect. Uh, just prompted anything without when you prompt chat GBT, what you get is a language in, language out.
So, very typical stuff. Now, well, we've built a system that allows you to get language in, useful economic output out, right? And that's a combination of large language models working together with agents, right? Uh what you're seeing here, ASI1 is a live product. You can view it or you can tune in by simply, you know, following the URL ASI1.
ai. This was made by FetchAI. It is a 39 billion billion parameter large language model that we trained in house, right? In combination with that, we have an agent hosting platform. Uh you can go ahead and my colleague was simp going to navigate to the agentverse quickly.
Yep. Uh what you're seeing here is the next component. This is the agentverse AI. This is an agent hosting platform. You can think of this as a port city, a hub where agents meet, right?
So we had the large language model, the agent verse AI, and then you can take a look at the documentation. What you're about to see is docs, and if you take part in our hackathon, you're going to be viewing a lot of this all the time. So what you have is three core components. The large language model, the agent versus new agents, and if you can just scroll it, this pick is everything you need to know in one picture. If I had to show you a slide, this this is it.
uh what you're seeing is models working together handinhand with the agent framework and the foundational layer the web three bit so to say as a chain which hosts a very key smart contract we have one smart contract we call it the almanac smart contract you can think of it as a phone book for agents so this the micro agent framework that you saw is free making an agent is free the framework is open source it costs you nothing but if you'd like and to make an agent discoverable to other people or to other agents, you're going to have to register on that smart contract. And that registration happens periodically to ensure livveness and to ensure that the agents that populate that contract are relevant and actionable. And so that's it. That is the entire system. You have a large language model coupled with agents, coupled with a hosting platform.
Why should you care about this at all? Well, because this system allows you to speak any human intent into existence. So if you'd like to execute a DY strategy, your typical route right now is find a wallet, download it, create it or import it, then uh connect it with any protocol of your choosing and manually do everything including obtaining the insight. And what you have here is a model capable of giving you actionable, relevant insight, not financial advice, just disclaiming it, but it can give you financial advice if needed. Um, it's also able to execute on your behalf.
So, you can ask the model to create any conceivable D5 strategy that you think has impact that that fits within your parameters and it will do it and then it will do it live in front of you, right? And if there's no agent that is capable of broadcasting the transactions or staking or compounding or executing the arbitrage, whatever the strategy is, then that's an opportunity for developers. So we currently have about if you can if my colleague over here can yeah if you as you can see top left corner we have about 2.7 million agents a lot of them are simply rappers of existing hugging face model. So in reality there's far less.
There's about 300,000 registered agents. There's massive opportunity for devs to pick up a very developer friendly framework, the U agents framework that has a depth and breadth of engineering rigor behind it and that has extremely refined documentation. It will take you between two and three hours to start from not knowing about it to being basically an expert. Uh you have a lot of examples to follow. If you're interested in making any conceivable use case, you can do it.
I I specifically use the term any conceivable use case because there are no limitations. And so if you'd like if you'd like to make an agent that preserves coral life, something abstract, you can do it. If you like a DeFi agent, if you'd like to make an agent that aggregates housing data and helps people pick out the best house, you can do that. It does not have to be web 3. In fact, web 4, web 7, it does not matter.
What the point is that humans have problems and we express our intents to solve them. And we've created an ecosystem that allows you to map out a problem, leverage pre-existing infrastructure, create an agent, host it, and any user that has that problem will find your agent. We currently have about 100,000 users, give or take on ASI1. So when you deploy that agent, you immediately have 100,000 users. As long as that agent is relevant for them, right?
So I'll pause here, right? by show of tail. Who got anything from what I've so far? Is there any clarity? Right, let me demonstrate.
A few a few heads are nodding, but let me demonstrate. Let's go ahead and prompt ASI mini life. Let's go ahead and prompt. So, I've actually
I prompted for the transportation in London, but this time, you know, the is just up. So, we cannot call the agent that's going to give us live information.
Perfect. So, we've given it a prompt and the prompt is uh given the current transportation information um find me transport information in London. That's the prompt. The intent is I want to travel in London. Help me be informed.
What you have right now is a generic large language model response. The links might not be real. The information might be hallucinated. That's because we haven't turn on the agent search feature. Now, let's go ahead and turn the agent search feature.
And let's uh let's just give it a find me an agent. Start with a find me an agent because you know so as with all things LLMs a little bit of prompt engineering massaging is needed to ensure that the model is able to obtain all the relevant context search for the agent and what you're about to see is the model started reasoning. We're going to give it 10 to 20 seconds. And if you're able to see the reasoning loop over there in the third sentence, it says because I have access to fetch AI's agentverse, I can search for specialized agents. So any agent you create and deploy immediately becomes a micros service.
That's very valuable. You can think of yourselves as micros service creators and it is very accessible. There's no the large language model also codes. It's a coding is language itself. So if you don't know Python, it's not a problem because syntax is a non-issue.
The model can create an agent and deploy it on your behalf. You just need to be able to document the problem well enough that the core task here is understanding what are high impact actionable problems in the real world, translating them into an engineering problem and making an agent that is able to credibly resolve that. Everything else is just a matter of submitting it to the agent versus in this case hosting it or if it's locally hosted. We have a message relaying system that allows your agent hosted anywhere. So if you'd like to host them in um secured enclaves like trusted execution environments, you can do that.
You can host your agent anywhere. So just to recap everything, we've now seen two responses. One is a large language model without agent search. And over there is the second one. You can just scroll to the bottom over o over there is the second response with agent search and you can see that there's five agents.
Can you just highlight them? Uh somebody created a transport for London agent and because moving around as London is relevant to me I'm now able to get real time data about what's happening in London's metro from while I'm in cans while prompting an AI model. You can do the same for any task. If you need help with your logistics cargo, if you need help buying a house, if you need help with advanced research and figuring out how proton f how protein folding works, you can do that. So there's no limit on the use cases here.
Why is that? Well, it is because AI is general purpose tech. It's like electricity. You can use electricity to well light a light bulb but you can also use electricity to uh drive a car. You can use electricity to uh create magnetic fields which in turn create levitating trains right mag lev technology.
So as a consequence of the general purpose tech that AI is it is up to us to figure out how to apply it. What we've created is extremely intuitive infrastructure and a large language model that you can use for practically no cost at all. And then we've now seen the large language model without agents with agents. And now let me just give you another example. It is what 15 I basically have seven minutes to wrap up.
So let's make the example relevant. Uh just ask for
Oh chat. Yeah. So um EV electronic vehicles are very popular. So, let's go ahead and give it uh um tell it what time it is and ask it to find live currently available Tesla Model 3 compatible EV chargers in cans with a 5 mile radius
Tesla Model 3 or any other Tesla doesn't matter. Tesla Model 3 EV charging with a fast charger at 1513 on the 4th of July in Kansas. you're about. Okay. So, we're going to give it 20 to 10 20 to 30 seconds to reason, right?
And what you're about to see is normally it doesn't have access to real life data like this. The today's date, today's type, right? But because we have an a EV charging station that's created and it is capable of aggregating between all of these agents, what you're about to see is uh just giving a little bit of a moment Okay. Find me an agent. It's It's still reasoning.
Still reasoning. Yeah. And there you go. Now, if you click on the EV charging recommendation agent once it's done, you know, let's uh we just give it a moment. Okay.
You can chat with the agent itself, right? So I if you're a user and you find an agent that's relevant for you that any developer on the planet made, you can directly chat with that agent as well. If you're a developer and you host your agent here and any user finds this agent useful and impactful, they can chat with your agent. And so it's not that you constantly need to query between a it's not that you constantly need to ask from within a search from within a pool of agents. people can habituate themselves to your agent if it's uh credible and worthy and if it actually resolves a known painoint right and the final bit is that so we've now we've spoken about the interplay between um humans and large language models and AI agents but other AIs can also use this right and so what you're seeing here is something as simple as a as a large language model becoming effectively a compound AI system you have the base model On top of it, you have the framework.
On top of it, you have um the agent verse and because the various agents can also be connected with other databases, real time information sources such as weather indicators for example, they can be connected with practically any other tool can be called right in so-called um kind of retrieval augmented generation architectures. You effectively have access to the world's digital tools, right? And what this allows you to do is find any product or service or tool or any information from within all existing products or services or tools. This is search without the ads, without the digital noisy footprint. If you have a relevant specific ask, as long as you're able to express it in language, you will find it.
As long as there's an a agent that exists that is capable of resolving that, you will be able to interact with the agent. And if no such agent exists, that's the opportunity for you guys as devs to make that agent. And so you effectively have a little bit of a two-sided marketplace. But uh on the one hand, you have users which need clarity and you know insight and on the other hand you have developers which need users and we're kind of bringing everyone together. Now uh I have uh I don't really have a lot of time left.
So um I'll pause here. I'll pause here uh and start a slightly more practical kind of conversation. Um we have we're organizing a hackathon. The prize pool is 10K. First place gets 5K.
Second place gets 3K. Third place gets 2K. This is the tip of the iceberg. It's not about this particular hackathon and you winning some arbitrary amount of fiat. It's about this is a highly soughtafter skill.
Being an agent developer is what being a solidity smart contract developer was back in 2016. People need this. People companies I'm not going to name name drop but companies from including from our partners said need credible AI agent developers. So this is a viable career path that is extremely high impact that allows you to be to to maintain incredible degrees of autonomy while also working at the edge of innovation. A lot of these agents are incredibly fun to make as well.
So, it's uh there's a sense of adventure here because you are genuinely pushing every conceivable limit. Now, if you'd like to take part in the hackathon, as part of that, I want you to know that in the ETH global discord, you're going to have 247 support. You are allowed to VIP code. Not only but we actually give you a readme that specifically optimizes the output of the large language model such that it makes the agent creation process easier. That is because syntax programming knowledge is no longer a bottleneck.
Finding credible problems and constructing credible solutions is what we're after and making them agentic. And so I'll pause here. U if you have any questions you are now is the time. Yes, sir.
I can link to Discord. Yes. Yeah. The Well, yes. Yeah.
Yes. Yes. So, we we we'll publish it I'll publish it live on the Fetchi X. How about that? This is the most centralized, easiest place to make it.
Publishing it live on the Fetchi X. And we have a booth over there. I'll make sure that you see the link live in the Discord. And then if you're if you're if you're hacker, simply tune in to the ETH's partner page and you'll see the Discord link there as well. You will be able to tune in.
And so uh once you tune in, scroll down to the artificial super intelligence alliance page and you'll see me there live, you'll see my colleagues there live. 247 support, juicy prices, viable career path, light work, especially if you're creative. None of this should deter you, right? you are all capable of basically excelling at making these agents. And um this concludes my 20-minute rant.
And if you want to go for resources, just go to our docs. Everything is there. It's the Innovation Lab FetchAI docs. I think it should be linked in the hackpack or something. And you got this the GitHub repo as well as Innovation Lab examples at Fetch AI.
So you're going to find the cursor rules. If you want to do some vibe coding with that with the help of cursor, you can do that. So it gets information from the relevant sources and you have quite a few examples there. So you're it's easy to get started, you know. [Music]
Automatic transcript — names and jargon may be misspelled.