AI x Ethereum: where we are now

Transcript

[music] Welcome to Devcon. It's good to have a good crew here. Do I see some Bhutan guys in here? All right. Shout out. What up? Welcome. We went to Bhutan a couple times. I think Madrul's here. You ran the last uh Bhutan camp, too. But it's simpler similar to this where we get a bunch of people in a room and teach them how to build on Ethereum. And uh that's changed a lot. And we're going to talk about that today. how AI has kind of changed the process of building [clears throat] and the process of learning and the process of learning how to build. Uh but I'm Austin Griffith.

I'm the founder I guess of the build guild. It's a big group of people. Um we started four or five years ago probably. I started the build guild by deploying a smart contract that allowed someone to Mr. DU I think was the first person to withdraw funds for their work and uh you know for better or worse lots of people took advantage some people didn't we had a lot of people stick around that were good we uh a lot of those streaming smart contracts led to really good contributors uh withdrawing and building and a lot of those contributors then ended up joining the build guild and that's the group here. So, the Build Guild is now a much smaller

organization. Uh, it in its probably when we really hit our stride a couple years ago, we were building speedrun Ethereum. And we'll get into that. >> Oh, yeah. This is not too loud or >> Okay. Okay. >> Check. Check. Oh, wow. Here we go. Okay. Yeah. So, uh, the build guild uh, in its heyday was building a lot of educational content and we were building lots of things. Scaffold ETH was our our main tooling and speedrun Ethereum was our main education. You could go as a builder and learn how to build on Ethereum uh going to speedrun ethereum.com. You can still do that today and we're going to talk about that a little bit more throughout the day and throughout the sessions. Uh but that's a

quick intro into the build guild. We're a DAO focused on builder onboarding and builder education. And this this is my watermark. This big messy thing here. This shows that I used my fingers to make this. And I wanted to make sure I had that there cuz basically I give AI everything and just let it one-shot all of my things. Uh it ordered me a car from the airport and I just trusted it and the car showed up. Okay, so that leads to my first question. How many people here are still looking at the code?

Like 25% maybe. And then how many people are writing the code by hand still? Okay. I It's It's more of like uh an artistic thing at this point, right? Like the guy making pottery in his basement. Like you don't you don't need to make your own bowls anymore, but some people still enjoy doing it. And I think that some people still love the syntax of code and coding. Uh but for me, it was more of a means to an end. I wanted to build a thing and put it in someone's hand and have them use it. And so, uh, now more than ever, it's idea guy summer. I can just come up with ideas and ship them left and right. And, uh, I want you to be able to do that, too. So, what what I've been building, uh, is at

the start of this year, I created, uh, well, it was right about the time that OpenClaw came out. I created my own bot called Claudebot ATG and I gave it a wallet and I gave it all the tools that we had been building at the build guild. I gave it speedrun Ethereum. I gave it scaffold ETH. I gave it a wallet. I gave it a GitHub. I gave it a Twitter account. I put money into it and it went and used that money and did a bunch of things. And uh kind of in the weird corners of the internet, uh someone deployed a token for my bot and pointed the fees at my bot and that token ripped to like 40 million. And so then my bot

had like hundreds of thousands of dollars in its wallet and it wrote a vesting contract for liquidity and it wrote uh an incinerator to burn the token. But it also wrote a whole bunch of interesting things kind of following uh some of Vitalik's tweet. He he talked about how uh Dowo participation is really hard to get. And so uh we built this app that was called Larve.ai where the token holders would stake their tokens and they'd earn conviction and then with that conviction they would get an LLM that learns from them. And so everyone gets their own personal LLM on the website for staking their token and

it learns from you and then can vote. So then when we wanted to hold votes, we would get 100% DAO participation because we could be like, hey, we're thinking about doing this thing with the token and then everyone's LLM could come in and vote. Uh then we made a ZK LLM API and you know that was just a research paper. We shipped it in a week and then the EF shipped it in like six months. But most certainly the ES version is much better. But my bot built one. Uh, leftclaw services is like a whole dev shop in a box that my bot runs. So, um, it can ship apps. It can audit apps. No one really used it until I meme'd $1 audit. And you can go to $1 audit.com

and you can pay one USDC and uh, I have a fleet of containers that will audit your contract. And the the point here is that it's a lot cheaper, right? Like when you think of an audit, you think of 10 grand, 20 grand just to get the initial upfront feedback on your contract. And so that was the point of $1 audit was to mimetically show that this is not the same thing as it was two years ago. This you're you're going to get the same quality that you would get from a $20,000 initial first pass audit for a dollar from some bots. And uh that has shifted the model of how we do things and it should almost be part of

your CI/CD pipeline. And $1 audits done like seven or 800 audits now uh fully automated. And wedgie.dev is the next thing. Uh we're going to do a Wedgie uh workshop on Thursday. Thursday morning. Come find me if you want to build your own hardware wallet. We've seen a lot of hardware wallets uh get hacked. uh supply chain issues, uh leaked address books, lots of problems with hardware wallets these days. Uh and so one of my proposed solutions is a DIY hardware wallet where you can go to any electronics store and basically order these parts. You can 3D print it and

then you can have a signer on an Osis safe. And if you find us on Thursday, we will be building these. And if you want to build your own, you can come find us on Thursday and build your own. And then uh on top of that is instant wallet. So I'm building a wallet with my bot that basically uses a wedgie. You take your hardware wallet, you plug it into your phone, and it can sign big transactions, but you can also send instantly within some spending limit. Again, all of this is built by my bot. I don't even look at the code. I haven't looked at the code for years now. Uh so we're shipping like crazy. Uh and then the other thing I have is slot computer. Uh I do a lot of interviews sort of uh

conversations around crypto and AI with a lot of creators and uh I spend a lot of time asking them about their AI psychosis and how they build and what tools they use. So check out slop.computer computer if you're interested in that intersection of crypto and AI and then ET skills and ETH evals. All of these things were started uh with me and my bot building and they kind of start as prototypes and then I hand them off to the build guild and they take over and do a much better job with it. Uh but those those prototypes are getting bigger better and better now with AI. You'll you'll notice that you can in one slashgoal, one session, you can one shot, you know, you can get 80%

of the way there, but that last 20% is getting harder and harder to do. Okay. So, uh what still needs you, right? I I think what it really comes down to is understanding what you actually build on Ethereum, understanding uh how the uh incentives are set up, doing the incentive design, and does the the critical thing though when you're building something, and this is one of the pieces of AI psychosis I would find when I worked with other people is they were, you know, I created 40 gigabytes of of information or I made this giant dashboard and it shows all these cool graphs. s and the question is like well who's [ __ ] using it like who and if the the answer is zero right

so it comes down to who's actually using it right the the true the true utility of something comes down to if people are actually using it or not and there's no better time than now to be a solo entrepreneur and get things in people's hands but there's never been a worse time in history to be a junior developer and so the the difference between those two is is really the way you think about things and the way you approach things. And so I would urge you to ship things and build things for people and get things in people's hands. And it really comes down to like taste and just distribution. Those are kind of like the big hard things now. Maybe they were the hard things all along, but us nerds like

building and so we uh would focus on the tech the whole time and maybe forget about uh some of the taste and distribution things down the road. Okay. So how is buildg guild adopting to all this adapting to all this? So with AI taking over, you can now oneshot apps, right? So uh you can also learn in one shot, right? with speedrun Ethereum. You used to come to speedrun ethereum.com and you used to spend I don't know weeks, months going through all of speedrun Ethereum and now it's more like you can get the knowledge, you can understand how you build on Ethereum and what kinds of things you build on Ethereum and what the gotchas are about building on Ethereum very quickly with this sort of like Socratic tutor that can take you

through but still ask you the questions as you go. There's there's basically like a slash start. You pull down the repo and you just give your agent the the challenge. There's probably seven or eight challenges and you just do a slash start and it knows everything about it already and you basically have a teacher in a box that will teach you how to build on Ethereum. So Pablo will be doing this first workshop at 12:45 and it'll be on speedrun Ethereum and the learning lab that we've been building at BuildG Guild. And then uh building on Ethereum with agents will be Shiv at 115 and he's going to take you through uh probably a lot of scaffold ETH is the

ETH skills and scaffold ETH will be building things with with the tooling kind of how the tools have changed and how you ship things quickly now. Uh I I give my agent ETH skills and let it rip and it can build lots of things. And that's kind of how we evaluated ETH skills in the beginning was give the agent eat skills and have it build something and see how well it does and then adopt adapt from that. Uh but then we moved on to ET evals. Uh so it started with kind of two different things. One was just me evaluating the builds and seeing how they worked. Uh but also uh the build guild was doing build guild ETH evals of ETH skills build guild ETH skills evals I think is what it started as and it's like what

we're trying to test here is is the knowledge in ETH skills does it really need to be there? So if you're building a skill file the last thing you want is a bunch of unnecessary information inside of that skill file. If the LLM already knows it you're just polluting the context when you have all that extra information. So you want just the right amount of information to be in your skill file. So we have this eval set that's constantly interrogating vanilla LLMs to make sure this information is the information you need. And Renat will go into this a lot more and say a lot smarter things than I am saying right now, but that'll be at 3:45.

Uh and then the capture the flag bring by ya, right? Bring your own AI edition. Uh, the capture the flag is obviously dead at this point, but long live the capture the flag. We'll have a challenge where you will use your own AI to take on a similar capture the flag that we did in Thailand, a similar capture the flag that we did in Argentina. Uh, but this one is I I would say like the one in Argentina was kind of a hybrid where we knew AI would be used, but we knew that like humans would still be doing it. In Thailand, it was all humans, right? There was no It's weird to think about those days, but it was only two years ago in Thailand. And it was like a

room of folks and a big screen and [snorts] flags getting captured, but they were doing it with their fingers and their brains, which is crazy to think about now. Uh, but you will still need to use your brain, but it will be mostly uh meat proxying your best AI through these challenges. And there's $5,000 of cash prizes at the end. So bring your own AI at 3:15 with Carlos and Damu. Okay. And then we're going to end the day with a kind of EF dev tooling roundtableish kind of thing. We don't know exactly how this is going to work, but we'll probably just get a bunch of dev tool builders and kind of chat in

here. Uh yeah, so find me this week. Uh I'll be in here later today doing that DevTool roundup. Uh on Wednesday I'm going to be doing a talk called Ethereum Transit Authority. It basically I containerized a browser and then ran a bunch of transactions on a bunch of different wallets and recorded everything that leaks when you do that and then talked about what it would take to basically have those wallets be fully decentralized and all the different services you would have to run locally. Uh then Thursday will be the wedgie.

We'll have uh the DIY hardware wallet where you can come in and build one. We have a hundred of the parts and all of the stuff 3D printed so you can come snap your own hardware wallet together and make your own signer on anosis safe. Uh and then yeah, I'll be h helping hackers at E Global. Uh yeah, then I have a talk on Friday at 12:30 about teaching your agents. Uh teaching an agent then letting it rip, I guess, is the name of it. I don't know. My my agent wrote that. Uh but it basically how I build on Ethereum will be a more in-depth talk and then I'm I'm doing some clear signing stuff on Saturday. Uh but that's it, I think. Uh

is it Pablo, you're you're up next? Is that the plan? Yeah. So I'm Austin Griffith on Twitter and Telegram. reach out if you have any questions. Uh the Build Guild will be here all day in this room building and teaching. Thank you for being here and if you have any questions, just let us know. We're here to help. All right. Thank you. All right. Thanks, Austin. Uh so if you didn't know this yet, all the uh sessions have this ask a question button.

So uh if you have any questions for this session now I think we still have time for that and you can go to the page of the session and ask the question there and this is going to be for every single session here at uh Devcon. Yeah. So definitely this topic of the cross-section of AI and Ethereum is uh very hot. It was in Vitalik's talk as as well. Obviously, nobody can avoid this uh new capability and uh yeah, I I also kind of agree with your conclusion that it's the the taste and the distribution which uh you have to kind of focus on.

It's more like all the easy stuff is done now. So, as humans, we have to think about the harder things and taste is like really really hard to cultivate and like grow. All right, let's see if there are any questions. >> All right. So, I can also ask them from here. Want to come up here with the mic? So, what if an agent start writing its own code and try to convince human decisions?

>> Well, I mean it should be, right? Like what if an agent start writing its own code? Yeah. I mean like that's it's coming, right? Like we have to write good guard rails. When it comes to good guard rails though, I think protecting ourselves from some of these things are interesting. And at that crypto and AI intersection, we can talk about the difference between don't be evil and can't be evil. And I think that you can build a skill file, you can write an MD file, you can have a brain for your agent that says, "Don't leak my private key." You can have a very good LLM that's not supposed to leak your private key, but it will still leak your private

key. So, I think that like there are certain things that we never want an agent to do, and those are the things that we need to lean back on on blockchain and crypto for for the can't be evil side. So, I think I think those things are coming. I think agents will be making decisions and trying to convince us that one thing or another. My my agent convinced me that I should take an earlier bus to the airport and uh the bus actually broke down and and we had to get a new bus and I was thinking like that turd making me ride on this early bus. But then I was thinking what if I was taking the late bus and the late bus was breaking down then I'd be in big trouble. So at the in the end the agent was right I think.

>> Okay. So it seems you're fine giving trusting the agent. So there's another question also about trusting the agents. Uh uh uh are agents powerful and smart enough to wode solidity contracts as well. >> Yeah, I think we're we're getting there. Right. uh that agent that I had uh in February and January was writing smart contracts that had hundreds of thousands of dollars in them. That's probably not a great idea. But we're getting to that point, right? We're we're getting to the point where it's I think when you pair it with formal verification, it's very very powerful. But we're getting to the point where agents can build very

quickly and then tests and checks and and $1 audits can do like that second pass to make sure you you've got it right. Um it I think this is like objective though or this is subjective at this point, right? We're in the transition period and it's sort of like if you're willing to let the agent write the smart contract, it probably can, but you need to do lots of auditing. Uh I do, but not everybody does. But we will get to a point eventually where it's not a good idea to have a human write the smart contract. Like you're going to get to the point where like you need the agent to write the smart contract to make sure it's safe. And we're not there yet, but I can imagine a

time where we will be there soon. >> All right. And the capture the flag is about reviewing solidity contracts today. basically. >> Yeah, there's there's uh it's sort of like that. It's more like there's going to be these challenges where you can imagine you have to have it's let let's dig into the solidity for a second. Like you have to have a couple of bytes of an address sign something. And to do that, you have to mine an address to get the bytes right to have it sign. That's just like a random example of something, but it's kind of like these twisted challenges that these two put together that was a lot harder a few years ago.

Uh, but now it's sort of like trying I think it's trying to be hard for even the bots, right? like you figure out like what can a bot do and what can't a bot do and let's like get to the edge of that and do something a little bit you guys used to be trying to build something that was kind of twisted for humans and now you're kind of trying to build something that's a little twisted for bots but the bots still like rip through it right yeah >> so it's sort of like solidity challenges but not reviewing the contracts more like there's a there's a tough function there's there's some tough solidity code here that you need to figure out what it does and then have your bot execute some function with something just right to get in. Think of it like picking a lock.

>> Quick check for the room. How many of you know what solidity is? >> Good. >> So most Yes. So that's the language you use for writing smart contracts on on the Ethereum virtual machine because there's a limited kind of vocabulary what you can do in the virtual machine and and that's you need a custom language that targets that. Uh then there's the question that uh for developers who are now like learning how to build on Ethereum, what would you recommend? How do they start their journey? >> That's that's a great question and that's exactly what we're here to answer. If you are a uh technical person or you are a good sloator, you are in the right place if you want to learn how

to build on Ethereum. That's what we're going to be doing today. Uh starting with Pablo with the learning lab and speedrun Ethereum. If you're at home and you're wondering how to build on Ethereum, go to speedrunthereum.com. It's basically the TLDDR is you need to learn all the interesting things about mechanism design and smart contracts. And to do that, you need to build a bunch of things. And those bunch of things that you build are condensed into speedrun Ethereum. And it used to be you had to build them by hand, but now you can build them with this like guided agent that will take you through things.

And Pablo will get into that next. >> Yes. So Pablo will start in a couple of minutes. And that's especially for people who want to learn Ethereum and how to build on it. But since we still have a couple of minutes and there's questions, let's continue with them. >> Let's rip. So related question also is for the hardware workshop. Do people need to uh do they need their own hardware to build at the hardware workshop? >> Nope. So the cool thing about this hardware is you can get it from almost like general electronic stores. It'd be a Raspberry Pi Pico. Uh there's like a a hat that's like the screen. It's a wave share LCD. And then there's a Trustm chip. And that trustm chip holds the

private key. It signs your transactions and it can't be removed from the chip and you lock it onto the chip. And all of those parts I shipped to Shiv's house and he brought them here. Where did Shiv go? I lost him. There he is. There he is. I I shipped them to Shiv's house and he's going to bring them in. And then some homies from ETH India and DevFolio 3D printed 100 cases. So, we will have everything you need. You just have to show up Thursday morning. I'm not even sure what room it's in, but come search my name on the schedule and you'll find it. It'll be Thursday morning for like an hour and a half. We'll have all the hardware, the case, and even like the software is at wedgie.dev and

instantwallet. I'll take you through all of that. So, nothing that you need. You You'll have to wedge some wires in. The hardest part is, and this is why it's called the wedgie, is you take that Trust M chip and there's no soldering or anything, but there's four wires that come off of it, and you have to put them into the header and then squeeze the Raspberry Pi on top of it. So, you kind of wedge everything together, and that's where it gets its wedgie name. Cool. Uh, last question. So, can you tell a little bit more about the capture the flag competition? So, it's going to be later in this afternoon, but people are already asking about it. Maybe they want to prepare for it. So, like how does it work? What are the prices? And

are you are you need to compete as an individual or can you be a team? And also like what's the role of AI in that competition? >> Yes. So, it's bring your own AI. I think you can have a team, right? No team. Okay. Whatever you want. If you want to team up, team up. If you want to just be you and your agent, but it's bring your own. Is it 12 challenges again this year? One hour, 12 challenges. Bring your own AI, let it rip. Like very few rules in terms of what you're allowed to do. Like use all resources at your availability and let it rip. And there will be $5,000 of prizes, $1,000 in Aztec tokens, and $4,000 in probably

ETH. Probably out of my pocket. and then we'll try to ask the EF to re reimburse us. [laughter] That will be at 3:15, I think. 3:15 in this room. Bring your own AI. Thank you for the question back there. And uh don't burn too many tokens unless you're very certain you're going to win the competition. All right, let's cut it there. Thank you Austin. [music]