"Deep Funding": Scaling Funding & Governance with Network Intelligence – San Francisco 2025

excited to have all of you all here today and really honored to be moderating a panel with each of you um all about deep funding. And so uh for anyone new and who's joining us and not familiar with deep funding, it's really risen within the past few months and as this new mechanism at really using AI um to have new ways to fund public goods. And I'm excited to be here with these four thought leaders um and experts truly experts in the space to dive in to all of this. Um would before we jump into the panel would love to just do a quick intro for each of you. If you could just say your name, what you're working on and the number one most important principle to keep in mind uh when it comes to anything deep funding related.

Where should we start? You. Okay. Hey, I'm Kevin. I guess I'm mostly known for founding Gitcoin which has funded $69 million worth of open source software and public goods in web 3 and I'm now working on a uh project called Alo.

Capital which is all about studying different ways of doing onchain capital allocation and that's how we've been supporting the deep funding project. My principle is that you must ship and iterate always. It's all shipping shipping and iterating. My name is Raymond Chang. Um I'm one of the co-founders of Open Source Observer.

Um, we're a public goods data infrastructure. So, we measure the value chain and supply chain of open source software and the impact that it has on various ecosystems including many of the ecosystems here. Um, and you know, we also help power uh some of the data for the deep funding contest. So, we're excited to be up here. Um in terms of one principle for machine learning um I would say in addition to ship and iterate and learn uh which I think is fundamental to machine learning um I would say keep in mind what your optimization function is.

Um I think uh there is a broad generalization that AI I think solves our all of our problems and I think it can uh under certain instances if we think about how we design it. Hi James. I work at Octant. It's a public goods funding platform that launched back in late 2023. Uh the big idea behind this is uh running grants programs on sustainability.

So what we did was we took a big treasury that was sitting on the sidelines and deployed that into a yield generating strategy. Uh and we have run all of our grant rounds simply from the yield. So having touched a dollar of the principal capital. Um so now what we're doing is we're creating the connective tissue so that other ecosystems can replicate this model um funding their growth sustainably. And the the kicker with that is we're asking each of the ecosystems that are using this tool um to kick a small portion of the yield that they generate to funding base layer Ethereum public goods.

Um so kind of dingling this ecosystem growth tool as the carrot and then getting them to contribute to some of the infrastructure that we're all critically relying upon if we're building on Ethereum. Um really excited to be on this panel because we're exploring deep funding as that uh mechanism for uh funding the sustainability or uh distributing grants through the sustainability pool. And I guess my uh the thing that I would would um focus on I guess for for the panel would probably be keeping the qualitative aspects of of uh how we're distributing grants like at the at the forefront of our thinking. Uh hey guys, my name is Dvanch and I'm leading the AI the intersection of AI with public goods and governance at the Ethereum Foundation. We've seen a lot of um crypto people get into AI but not so much AI people into crypto.

So that's kind of one of my mandates. How do we get them interested in our world? Um kind of like one principle I would say is uh only crying babies get fed which is kind of the usual principle in all of these funding mechanisms that you actually have to apply for grants and like do a lot of crying before you get money and uh a lot of the best teams don't want to do that. So actually you exclude like giving money to like your best teams. So how do we actually solve that?

And deep funding is one way of solving this issue. Only crying babies get fed. I love that. Getting a that should be the new uh motto on Gitcoins. Uh back at Gitcoins.

Thanks. Um but Gitcoin's going multi-mechanism. We're not just QF now. We feed all babies, not just crying babies. Feed old baby.

I mean that's a perfect segue into my next question I do have is what is broken with our current funding mechanisms and how does deep funding or other types of mechanisms and capital allocation mechanisms address those? Do you want to kick us off? Sure. Um so I think that there's this really interesting design space of how do we fund public goods because the most effective ecosystems will fund their public goods things like open source software things like education community. Um but there's a big question of how do you do that?

Um and you know I think that the p person who sort of like figured out an answer to this actually is Satoshi Nakamoto who through proof of work created a cryptographic mechanism through which you could validate that you created a block a public good which is the next layer of a blockchain and then you get the block reward for it. Right? So um the reason Bitcoin exists is because of public goods funding of proof of work and um you know the question is how do we fund other public goods beyond just like blockchains and there's a whole design space here of people who are doing quadratic funding or retrofunding or cookie jars or continuous retrofunding. Um we came up with an idea this morning for uh configurable politicians and then we forked that to make uh bribable configurable politicians. So there's a whole design space here and there's going to be multiple ways that we fund public goods in the future.

Um, and they're all going to be serve different value niches than each other. So none of them is better than any other. It's like it's like art. It's like a you can't choose your favorite band because different genres of of music are not like comparable with each other in that way. But they all like fit different value niches.

And deep funding, what it's really good at is dependency funding. So the dependency tree of uh Ethereum is Go Ethereum and Prismatic Labs and Nimbus and all these other clients and what are the things that they depend on and making sure that those things are getting funded uh down the dependency tree. So I think that'll tee it up for the rest of the panel to talk about defunding. Yeah, 100%. I um I would say to to answer your question, I think one of the biggest blind spots right now in the public good space um is evaluation frameworks and especially machine learning.

This is going to be critically important when you think about what it is like as these uh algorithms are continuously getting better. What are we getting better towards and who's defining that and how do we think about that? I think there's been incredible experimentation already. I think this community is one of the best communities out there to think about new types of mechanisms. But collectively as an industry um I would say in crypto alone probably has distributed many billions of dollars already in public goods funding.

And compared to classical like you know classical economic policy or some other domain there just isn't as much rigorous study yet on what drives what outcomes what mechanisms are good for what purposes. Um it's one of the things that you know I I really respect some of the work that you've published Kevin and in the world of enumerating the different types of mechanisms and starting to think about on what under what conditions different mechanisms thrive but I think we need to get a lot more quantitative and I think that's one of the reasons why at open source observer we're trying to build this culture of a data science community to look and analyze um the different economic outcomes that we're actually producing from the economic policies that we are using here. Um it's one of the things that you know we're collaborating with funding the commons on um and hopefully uh initiating some new research in this space. Um but I think you know there's there's definitely a lot more to be to be improved there. Yeah.

I think uh another point to answer your question is um at least Republicans funding a lot of the the rounds that we see are community based and so there's a lot of different donors that are participating and I think donors distilling information and knowing like where is their dollar going to be able to drive most impact is really really challenging. Um we have 30 projects that participate in each of our rounds at a time and it's hard for our community to be able to distill information there. I can't even imagine when you have you know hundreds of projects or or even more. So, um I think AI can do a really great job at being able to uh summarize and and give you know key information to donors that uh you know are suffering from uh attention uh deficiencies. So, um yeah, we're really hopeful that uh something like that is going to be able to um support our community in the future.

But definitely a problem that we see right now. Um, yep. I think just building on that I see three main issues like if any of you are designing new funding mechanisms try solving any of these three issues and you made a big contribution. One is uh as you said like letting voters focus on quality over quantity that in a lot of the funding rounds I need to have an opinion on every single project taking part in the round and I don't want to have an opinion on every project. Um so actually like being able to just focus on five or 10 or 20 projects and AI can extend my reasoning to the entire set of 500 600 projects that would be like far preferable and that's what deep funding does which we'll get into quality over quantity.

The problems that deep funding doesn't solve is um the halo effect is one. Um halo effect is basically that because a project has some reputation from the past, jurors just like blindly give it a high score. And we saw this happen with one of the repos called web3js which uh is like a it's a very core thing of ethereum and it's done a lot of work in the past but it's been overtaken by vhim and one other repo ethersjs and but the jurors all gave it high marks and the ef did an investigation which found that less people were using it and it's been overtaken and they shut down actually in the process of deep funding and I've seen such high marks given to them in a jura script. uh halo effect is something we don't solve in deep funding too well since you are replicating human judgment and human judgment suffers from the halo effect. Um and the last problem is parts out.

This is actually a very nuanced issue which OSO has actually written a lot about. Parts out value is basically a term in the mechanics industry where if you sell the whole car you actually get less money than if you strip out the car and sell it each of its pieces. So we see a similar issue where in a gitcoin QF round or even in badge holder round in optimism if I actually divide my project into five like sub projects each of which take path I get way more money than if it was all just one project. So that's something to be seen if deep funding manages to solve or not because if you divide your repo are you going to get more there are more predictions of machine learning models but they might each get a less weight. So that's like to be seen.

So paths out value, halo effect, and quality over quantity for jurors are the three things. Yeah, you touched upon a ton of really good points there. And I think one of the things that we've really noticed with AI becoming a lot more prevalent in all of our uh onchain funding mechanisms is our decisions around the evaluation frameworks and what impact we actually want to fund are more crucial than ever before. Raymond, I do have a question for you with open-source observer. What are maybe the risks of being that sole entity that kind of does decide these impact metrics for retropg and how do your decisions um have drastic effects on who does or doesn't get funding?

Yeah, great great question. Um when we when we started working with optimism, the first thing I told them was I I asked them I asked them two questions. I was like uh first of all, are we doing pathological science here or are we doing real science? because there's a whole set of data companies that just exist to make charts look like this for whatever purpose and that's not what I wanted to do. The second thing I said was like I don't want to be your McKenzie like there is no desire here to be the authoritative source of what impact is.

I think impact fundamentally is subjective but can be there are aspects of it that can be measured. And so while we've been building this out and if if you're not familiar with the retrofunding experiment, I can try to draw a hand visual diagram, but it's basically, you know, there's uh the crypto economy is one of the most visible economies in the world. It's incredible. You we can see every way that a project is funded. We can see every single commit that a developer is pushing into GitHub.

We can see their entire dependency tree, which tools they use, which libraries they use, what packages as part of their software supply chain. And then we can see every single time you ship something. And for you have global visibility on chain. So for every single application, not only can I see all of your revenue for whatever application you are, I can see all of your users and how high quality they are and how engaged they are and everything. It's like it's the most incredibly rich economic data set ever.

That's the way that I pitch it when I try to get economists to come in and help us analyze this information. So what we do as OSO is we we vacuum up all this data into a giant data lake that's a public good. So we we release these public data sets for for anybody in the research community to use and then there's this data pipeline that transforms the data into a set of different models that basically get connected directly to funding rails which has delivered many millions tens of millions of um optimism token at this point. And one of the things that's incredible here is one, you see the entire provenence of how you are being rewarded and how you're being evaluated, but it's also highly flexible and configurable. When we first started, the first blog post we ever wrote was open source, open data, open infrastructure.

Again, we we have zero desire to be the arbiters of truth here. Um, I think one of the things that we're planning on shipping this quarter is, um, in addition to giving you access to all the public data sets, um, giving you the ability to take any model inside this pipeline, fork it, like view it, see how it like test it, play with it, fork it, make your own, um, version of it. If you don't like how user quality is evaluated or you don't like how application developers are being evaluated in terms of retention or other metrics, help us improve it. and then you have the ability to publish it right back into the pipeline and directly affect um how you know all of this uh money is being distributed to public goods. So it's an incredibly powerful flexible piece of infrastructure.

We scan hundreds of terabytes weekly. So it's it's an insane piece of infrastructure that we've built. Um but now the goal of the next quarter is to basically open it up to governance. And while it's always been the case that anybody could contribute, it it does require at the current point a lot of technical expertise to understand how to get in there. And you know, if you've never played with SQL Mesh before, if you've never played with DAX for some of these like, you know, really in-depth data engineering tools, what we're trying to do in the next quarter is try to lower that friction.

So anybody with a Python notebook, anybody in a Jupyter notebook can come in and publish a model into the pipeline. So um great question. I really appreciate that. Um, and I I I invite the broader data science community to help help us figure out how to measure impact. No, you're totally right.

This is one of the key stones behind this technology is we have one of the largest data sets about economic impact, about how to create new economic models that fund things that were never funded before. and how can we be louder about that and attract new people who don't know anything about blockchain maybe not aren't even interested in this to bring their expertise um I in and bring their expertise and help with the same problem um I do want to ask though with AI becoming a lot more prevalent and say AI just becomes so exceptional at finding what is the best way to genuinely fund public goods to I guess make number go up or whatever our ultimate impact metric is that we're tracking. Where do humans now come in in this process? Uh I would love to start off with you Aoki. You recently wrote your book, How to Dell.

I'll show it for you and talk a lot about human coordination. Um how does that look like in the future where AI becomes a lot stronger uh and can make better coordination decisions potentially than humans? Yeah, thanks. Um well uh with only three minutes left I think I'll be brief and just say it's an incredibly large design space. Five minutes left.

Cool. Still we'll make some space for Devon and James. But uh uh I'm really excited about Vitalik's vision of uh you know taking human judgment and distilling it. You know one of the problems that we have in the attention economy is getting voters to express their preferences. Hey Raymond, come vote for my proposal on Bitcoin Dow like every quarter for the rest of your life.

Like it's it's just like there's governance fatigue. let's just say it. And so by ingesting our preferences into AIs that can represent the distilled human judgment of the people around them, we can amplify the signal. And uh I think there's a lot of power in allocating resources more democratically with that increased context that is input into AI. And then I also think there's another really interesting thing in Dowels where you can have an AI that reads the governance forum and the discord and all the meetings for you and just tells you what happened also.

So it's about distilling signal and intelligence and feeding it to the resource allocation uh algorithms and to the people and I think it's going to scale our ability to coordinate 10fold. Yeah, it's almost like human superpowers, right? Rather than than just doing it all. That's a good meme. Yeah, we'll play with that.

Human superpowers. Um I know we only have a little bit of time left, but Devanch, we were earlier talking about uh dinner you went to last night with Vitalik where you were talking about some of the bottlenecks and pain points with deep funding. I would love if you could share maybe one of those uh with the audience and what the result was. Yeah. So maybe I'll talk about like his biggest concern which is the civil issue where technically as a model submitter in the deep funding challenge.

Um just like quick what deep funding is is we have 15,000 edges in the Ethereum dependency graph. So this is about 5,000 open-source repos that power Ethereum. And the challenge is to basically give a weight to each of those repos or the edges and then we put money in and it gets distributed to all of the repos according to the weights that the winning model chooses. So how do we choose the winning model? There's a jury which is going to give its preferences on maybe 500 of the repos or even less and we use that to select a winning AI model.

So the issue which Vitalik had which we managed to solve at dinner last night was um you could technically submit an infinite number of models as a machine learning contestant and one of those models is going to fit with the jury data. So how do we actually stop an infinite number and like people just like mass submitting models hoping that one of them is the winner and they get a lot of money for that. So uh kind of the two approaches were taking verified phone numbers which is what community notes does. community notes to be a contributor. You It's not just a phone number.

It's a phone number from a reputed carrier. So, that automatically limits it. And the second option is just charging $2 per submission, which would also fund our deep funding team, which is amazing. Woohoo. No, I love that a ton.

Um, and James, I did want to ask you about Octint V2 since I know that recently launched. and specifically when it comes around one of the things I really love about Octant is the model of an endowment fund and how we can have long-term games when we think about long-term uh long-term funding mechanisms. I'm curious with how you see AI and its role in infinite versus um finite games and how AI might push us more towards um infinite thinking or more towards short-term strategies. Yeah, we've on the team, we've been thinking about this a lot. Um because this pool that I'm talking about in which we're trying to entice a lot of these other Ethereum ecosystems to all contribute a little bit of their yield to to collectively funding uh what's critical for Ethereum.

Um defunding becomes a really interesting, you know, uh topic to explore there. But I think the concern and what I was trying to resolve in my mind for a while is how do you embed the the qualitative aspects to to what's really important in figuring out what should be funded? Um, I was actually talking with the CEO of uh, Metaculus here a little bit earlier and and he resolved this for me. If you don't know what Metaculus is, it's this amazing forecasting platform that um, has achieved some really really great results over the last nine years. And what he was telling me is that um, in order to uh, start to embed the qualitative side of things, what you can start to do is uh, put these inflection points in uh, into the models.

And so these are going to be things like um uh milestones that a project uh hits or you know some kind of partnership or um some kind of adoption maybe some like massive technological breakthrough uh along that project's trajectory. And so when you're able to essentially embed these inputs into um into the models, it actually can be be quite um quite good at predicting things. And so um that kind of resolves it for me and how we might be able to you know complement uh human input into scaling you know how these AI models are essentially going to be um deciding how funding is distributed and so pretty excited about that. Very exciting. And I know we're at time so I wanted to wrap it up with just one closing thought.

Um I as someone who leads developer relations at Cello a lot of what I think about is sustainable funding mechanisms for our builders. I do believe that builders are the blood behind Ethereum and what keeps everything moving and keeping them funded is uh super important to ensuring that this infinite garden can really grow. With that, is there one last word you would like to say to maybe any builder that happens to be watching this? Ship it. Yeah.

I would say make impact like don't focus so much on the metrics like just go it's standard OKR theory go achieve the greatness and you know the goal of these systems is to just reward you proportional to how well you know you're achieving that impact. Yeah. I think for us the story that we're trying to tell with what we're building is this idea of deents and dragons. So there's a bunch of dragons in the ecosystem sitting on piles of gold that aren't doing anything with it right now. So trying to get these dragons off their ass to deploy capital that's sitting on the sidelines and um you know start to fund things with that yield.

Uh and then there's a bunch of degenerate gamblers just trying to make a better financial world for themselves. So how we can start to turn them into uh more positive sum games and is what our goal anyway is for this uh story that we're telling. Yeah, it would be to like build in public and to keep tweet like a lot of builders think that I just need to ship my project and that's the most important thing. But Awoki is like a great marketer like he tweets all the time and that's a huge part of the impact he creates inspiring me and others to get into the field. So that's an equally important impact and that's how I got involved with deep funding and EF and all of that like by tweeting Vitalik saw the tweet messaged me and that's how I am where I am today.

So tweet publicly build in public. That's amazing. Super honored to be here with you all today and thank you all. Thank you. Thank you.

[Music]