GG23 AI Experiments with David DAO & Devansh Mehta

Transcript

Welcome to the Aloe.cap podcast. If you're just joining us, we're doing seasons. This season, we're summoning Aloe Capital, a new spinout from Gitcoin that is designed to build the capital allocation layer of the tokenized internet. We believe there's big upside here. Capital allocation is a multi-trillion dollar design space. VCs, governments, NOS's, Dows, and regular people all have to allocate capital. But it's also a place where we can start small. Every time you spend money, you are allocating capital. We are threading a through line partnering with it with builders, researchers, and funders to figure out what onchain capital allocation looks like and building it on chain. If you want to learn more, visit our website at alo.cap where you can download the Aloe books for free, play with some of our DAPs, join the Telegram, or participate in the

ecosystem. All right. Today I am playing a recording of a Twitter space that the Gitcoin account held last week in which we talk about some of the bold capital allocation experiments that are happening in GG23, Gitcoin's 23rd carnival of web 3 funding where we fund what matters to the Ethereum ecosystem. On this episode, we are going to play the Twitter space in which we go into two bold new experiments in AI capital allocation. In this episode, you will hear about those bold experiments and what makes them bold, why I'm excited about those, but also I encourage you to take a look at how Gitcoin is evolving from being just quadratic funding to being a multiple

mechanism shelling point for different builders in the space to do innovative ways of funding what matters in the Ethereum ecosystem. So if GG1 was all about quadratic funding and GG23 is quadratic funding, retrofunding, and a handful of bold experiments around AI capital allocation, I can't wait to see what GG24 and GG25 and the future of a pluralistic set of mechanisms for funding what matters in the Ethereum ecosystem looks like. And uh it's a great joy to be at the epicenter of making this all happen uh with Gitcoin and stewarding Gitcoin, but also Alo Capital is working to stimulate the multi-mechanism future of Gitcoin. So um I think you're really going to enjoy

this episode. Without further ado, I give you the Twitter space hosted by Matilda, the grants program lead at Gitcoin and then featuring Devon Meta and David Dao who are ecosystem builders who are working on the mechanisms. Enjoy it. [Music] GM, welcome, welcome. Maybe we can kick off while we are still waiting for a few more people to join us. And if you are part of the space, please like, share, retweet, whatever you got to do to get more people into the space where we're

going to be discussing all the experiments, well, some of the experiments that we ran alongside u Gigi. Yeah, super excited. Uh maybe we do a quick round of intros and then we can get started. Um for those of you who don't know who I am, my name is Matilda. I'm grants program lead at Gateway. Um yeah, super excited for this conversation today. David, I'll I'll hand it over to you next. Hi, thanks a lot for the invite. I'm David Dao. It's my real name. Um, and I'm the chief scientist of Gainforest now. And also, um, we are running right now a really cool deep cuff experiment with Eval Science. And I think that's something we're going to tell you about later. And maybe I give it to Carl. Thank you, David. My name is Carl. I work with Al

Capital as a developer and also working with David on the really cool new experiment running on DG23. So yeah and thanks everyone and sending it to who's next. Uh why don't we hand it over to Devanch next? Um hey there. Uh, great to be here. And with along with Bill, we did like a cool predictive funding challenge where AI models had to guess how much money projects would get in the round before it even started. And it's been really cool just seeing the error rate and how close uh model submissions were actually to kind of calculating. So I'll hand it over to Bill next. Yeah. Excellent. And hi everyone. Yeah, this is Bill from from

Pal. uh we are we're hosting this competition on you know on predicting fun you know predicting how much how much funds you know the projects will receive in G23. Uh it's a yeah it's a very fun experience. We you know we have the I I think we have a dozens of models submitted and it's fun to see how they how their performance changes in real time as the fun started. Amazing. Super excited to dig into all of that. D, I'll hand it over to you next. Nice. Hey.

Hey. My name is D. I'm the grants program coordinator at Gitcoin, and I'm just super excited to dive in with the audience and kind of learn more about all of these incredible experiments that are happening. Amazing. And last, but definitely not least, uh, Wookie. Hey, what's up? I'm Kevin Walke. I'm one of the original co-founders of Gitcoin, and eight years later, we're still cooking, still trying to figure out how to fund what matters. It's been an adventure over 23 rounds and Gitcoin is evolving once more and I've been really excited to see as we go from quadratic funding to multi- mechanism what kind of cool signal generation we can create and what kind of cool capital allocation we can create to solve Ethereum's biggest problems and to fund what matters in the

year 2025. So let's get into it. Amazing. Let us definitely get into it. Okay. So first of all I just wanted to give a quick t on what GD23 you kind of preempted this as well um Aoki so thank you for that GG23 was a really like the whole of GG23 was actually an experiment you know we moved from purely quadratic funding in our own program to a multi-mechanism future within ours program and we saw also a ton of mechanisms coming out of the community rounds this round as well. I'm I'm so proud to see this round come to life.

It's been months and months of strategizing and figuring it all out and it's been a really big moment to actually just see it in you know you know come to life and um yeah super excited. One of the other things that we also really wanted to push forward was this air of experiments within within funding but also what kind of side quest experiments almost can we can we see come up and two that we're going to be discussing today is like our speakers already mentioned was the deep gov experiment as well as the predictive funding. This kind of ran side by side to both uh to the GD23 rounds which is super exciting. Yeah, and let's get into

it. I don't want to hear too much of my own voice in this in the space. So, definitely we'll open it up to the floor. So, I want to throw it over to predictive funding first. Devon, I'll hand this over to you. You know, can you just give us an overview of what was the goal around this challenge? Uh, this did come in as a GCP to our forum. Getcoin is also co-unding this. So, yeah, super excited to hear. Can you just give like an overview of what the goal was, how it's going um and then we can go from there? Yeah, so basically I got uh hired at the EF sometime around February in in early Feb to look at how AI can be used for capital allocation and kind of one

of the biggest challenges in using AI for allocation is accountability. That how do we even know that the AI is good? Are we just like yoloing the money into an AI which might not be very like it might not be the best AI? it might not perform the best. So kind of the like we've been taking like a data science competition approach something like Kaggle where you have ground truth data and um we see like we and many people submit an AI model and then we can actually rank the AI saying this is number one because it's most closely aligned with the ground truth data and this is number two and the first challenge we ran was based on past Gitcoin and Octant and Optimism and Open Collective funding data where we kind of

relied on an owner system that even though this data is available publicly like build a model that predicts the funding which projects would have got in the past and we will judge you based on your writeups in the forum. So we got around 21 good submissions there and like an error rate of 0.025 was like the winning model and this is going to be important and I'll come back to it. So and like the biggest feedback we got from model people was that we don't want jurors judging and giving prizes. We wanted to be purely based on the error rate and the leaderboard uh placement. So that's when we saw this kind of very brief window of time where applications ended and 36 hours later the GG23 round starts. So we basically um like in those 36-hour

window you can make a prediction of how much funding each project is going to get in the round and we just ran it like we did it for 36 hours and we got about 11 or 12 good submissions which is honestly more than I was surprising which was more than what I was expecting and currently like the QF payouts are still left so that's going to determine the final leaderboard placement but currently I think there's an error rate of around 0.034 034 or something. So the like when the past data was like fully known it was 0.026 here it's 0.034 and if you just guessed randomly it's like something like 0.045. So we've seen like some kind of things hopefully the error rate comes

down much lower and the kind of larger motivation and I'll end it here. The larger motivation is that if the error rate becomes low it basically means that uh we don't need to run these QF rounds anymore. We can just ask like if a new project comes and we want to figure out how much money to give this new project. we can just ask the top performing AI that how much would this project have got if it had taken part in the QF round and we would get like a decent approximation of that. So I think that's what kind of the larger like the longer excitement about it is and it's also just a really cool challenge for a lot of like AI people. So we've seen really good submissions and we're getting the community activated and involved. Um, so

it's also about how we make the crypto space cool for AI and not just the AI space cool for crypto which is what we are seeing right now. Yeah, super exciting. Bill, anything you want to add there? Sure. Sure. I think yeah, just probably I just want to add you know first add some like interesting things that happened in the preparation of this competition. I think you know we we started like actually quite late. So I think you know you I and the de had a meeting probably on Monday or something like that and then we have this competition out on Tuesday or something in 24 hours. That's definitely a a a good experience to have. Uh uh so but yeah but but we made it you know we got everything uh sorted out and started a

competition. uh as Denture said uh it's it's um unexpected to have so many to receive you know so many submissions in just a very short time window. So I guess one one thing I probably want to add is oh we also learned a lot of lessons right from doing this in such a short period of time. Of course there's some uh hiccups along the way. So I guess you know if there's another round if we're if we were to do it again we will definitely be more prepared and uh you know give a even better uh experience to the developers. Um but I think that's one thing I want to add and another thing is like probably people might be wondering right what's ultimate goal. So I think one goal is you know

what Danch just said right is to maybe up with the QF funding uh you know the the QF realm and then another thing I think probably the ultimate ultimate goal is as you know italic you know wrote in his post earlier which is to basically trying to scale human judgment right so if think about in the future where we have just so many things we want to fund or we want to support and then it's very hard to you know get people to vote any about on every one of them or it's even hard for people to discover all of them. So maybe you know this is where we can use uh AI to help right help people discover the projects better and also help them understand the project better and also on the you know

on the flip side maybe you know we we don't need to really you know go like 100% aligned with what the model predicts maybe we when we see like like the the discrepancy between what the model predicts and what people you know actually do maybe we can you know find some insights there as well you know to to maybe to figure out you know is there a like a hidden gem there that's you know kind of overlooked by people but then the the uh you know the AI picked up or maybe you know I AI you know saw something else in that project that's worth looking at you know maybe AI don't think that project should be funed that much uh then yeah then it's also interesting case to look at too so I

think yeah there many use cases you know or many things we can use AI in the future if we you know get this thing Right. Yeah, that's really really interesting. I love that piece around project discoverability as well because that's something that it's a lot of it's part of a lot of feedback that we've gotten, you know, with our QF rounds. I mean, if you just look at the OSS program, we usually have over 200 projects participating and especially if you know, if you're coming in brand new, it's project discoverability can be really tricky. Um, so I love any kind of other aspect of just increasing how we discover projects in a round. Awaki, I know you're super excited and like super bullish on all these types of things.

I'd love to hear your your feedback as well and how do you think this went, what you're excited about as well and then we can dive into deep a little bit as well. Yes, I mean I'm I'm really excited about Gitcoin going multimemechanism more broadly and uh you know I think the mission of Gitcoin is is about funding what matters and specifically funding what matters to the Ethereum ecosystem is sort of our core constituency and uh I think that we were solutionist for too long just trying to figure out a way that we could wedge quadratic funding into everything and um I think we that ran its course we found some great designs for quadratic funding and really helped solve quadratic funding's biggest problems through doing that. But now is the time to take a look, a fresh look at how capital

allocation can be done on chain in 2025. And AI is just wow. The combination of permissionless global programmable blockchains with large language models and commodified intelligence that can be allowing people to signal their preferences for where they want the money to go in a very minimalistic way or uh use using existing metadata exhaustion where they don't even have to put in new data and the AI can can come up with new ways uh to represent the people's will. I just think that that's such an incredible design space and I think of the predictive funding experiment that we did in GD23 as as just sort of the first pilot of uh how can we use this in order to do better

capital allocation and and honestly it gives me GG1 vibes. In January 2019 we did our first pilot of quadratic funding. We had no idea if it was going to work and we learned a lot by just doing the first case study and getting that empirical data of how it worked. And so I'm excited about the results. I think that this was this was definitely a successful experiment. And I think that the real question that I want to orient people towards is what do we do for GG24 knowing what we've learned uh in in GG23? What are the ways that we could remix these experiments or these mechanisms or extend them in such a way that we learn even more about how to solve these problems? So that's all I'll say for now. I'll probably have more comments after we get into the next experiment. If you're enjoying this

episode, please check out the remainder of the episode on the feed hosted at aloe.cap. You must be a aloe.cap patron NFT holder to hear the rest of the episode. Alo. Capitalap patron NFTs give you access to the Illuminati, which is a private community that has access to the Alo's evolution, novel capital allocation, intelligence, and research like this episode, like this podcast, and also research reports that we are publishing at all. and early access to builds and other innovations in the Aloe community. Thanks for listening. If you want to hear the rest of the episode, check it out at aloe.c capital. Peace and love.

[Music] [Music]