Deep Funding: Driving Rewards to Open Source Repos Using AI

Transcript

kind of refund dependencies of core repos and not just the repos themselves. Anyone can run a deep funding round and you know can have their dependency graph whereas deep funding gives to repos like we give money to nethermind for being an execution client or a consensus client. We are really focused more on how do we allocate this is the controversial part now you that's a perfect segue into the controversial ones. Anybody who's passionate about these repositories or like are familiar with the code are really welcome to be a jury [Music] member. Welcome to ecosystem project demo the podcast where we spotlight public good projects shaping Ethereum

ecosystem. This is episode 19 brought to you by EC Institute and I'm your host Puja Ranjan. Today we are diving into deep funding, a unique model that rewards open-source contributions using a market of AIS as engine and human as steering wheel. If you have seen Vitalic advocating for it online, you know it's worth paying attention to. In the next few minutes, we'll explore what makes deep funding stand out, how it works, and why it could play a big role in future of decentralized funding. Joining us to unpack it all are Danch Meta from Ethereum Foundation and Caes Rahan from

Alo Capital. Thank you both for joining us here today. Yep. Thanks for bringing us on, Puja. It's all my pleasure. Personally, I found the concept very interesting and it is new as well because it includes AI. So, I'm sure people like me are curious to understand its use cases and also how Defunding is implementing AI to help this decentralized ecosystem. But I feel like it is always interesting to know people behind the scene. So, for those unfamiliar with your work, may you please share about yourself and your journey in the ecosystem? Yeah. Okay, I can go. Thanks Puja for having us. Hi everyone, I am Sel. I started my journey with Greenpill Network two years back

which was a podcast that Kevin Awaki was running and then my goal with Greenpill Network was to enable local communities to talk about all the great tools and mechanisms that are being built in this space and how they can use it at the local level. Right? So that's how I started and then Gitcoin was the next obvious thing that I wanted to do because I wanted to better understand the mechanism public goods funding mechanism design space and there I first started with GG program where I was running bitcoin grants program and I was also a grantee with green pill network.

So I had kind of seen both the sides of the things where you know as a reviewer what this looks like and as a grantee what this looks like and uh that is what always interested me in this space to sort of further explore one what are the different ways and you can actually fund mechanisms because QF was sort of like dominating the space and then I ran easy RPGf where we decided to you know like optimism has pioneered retro funding and how we can enable any ecosystem to run retro rounds right these These have its own challenges that sort of further you know intrigued me to explore this space to see what are the other mechanisms that can you know sort of address some of these challenges and that's how I ended up in working with deep funding

and now I'm part of Alo Capital which is a spin out of Gitcoin which is like specifically focused on sort of exploring the design space of funding mechanisms. So yeah happy to be here and diving deep into deep funding. Thank you for having me Danch over to you. Yep. Yeah. That's a good introduction and hard act to follow up on. I mean I entered crypto in like 2021 and before that I was a journalist working in like in the central India regions where there's a civil war going on and kind of I entered crypto as part of like a funding model where a lot of these journalists were creating impact with their work like teachers getting reprimanded for not taking class or a

road getting built in a village and kind of the idea that got me into crypto was what if we could represent that impact as a unique digital object. project and then people can buy that to fund the future journalism work as well. So that got me into crypto and then I stumbled into Gitcoin grants when I met SEL as well in one of the green pill events and kind of we finally implemented that idea. We got like impact certificates and hyperserts for journalism live on voice deck and kind of I tweeted at the end of DevCon. So this is like maybe three or four months ago. I tweeted that I'm very interested in AI quantification of impact that we have all of these impact certificates from journalists.

how do we now quantify that impact using AI? And somehow Vitalic saw that tweet and then he kind of messaged me saying, "Hey, I've been interested in something similar." And then he lays out this like super sophisticated mechanism. And I think kind of we've all been thinking along the lines of kind of let's build a very good AI model which can then quantify impact. But the breakthrough insight that Vitalik provided was that let's not try to create our own one model to rule them all type of framework. Instead, let's create a competition for many people to submit models and then we select amongst the winning models. And that was quite like a breakthrough insight which I hadn't like kind of even thought of or

considered at that time. And kind of after we started deep funding, we said let's apply this to Ethereum core dependencies and that's where deep funding began. And um after that in kind of I got offered like a role to lead some of the AI and public goods work at the Ethereum Foundation based on the progress we made in deep funding. So it's more important now than ever to actually do something good. And kind of the one thing I'll say about my role at the EF is that we've seen a lot of crypto people get interested in AI and I think my overall mandate is how do we get AI people interested in crypto and having competitions like deep funding where AI people can actually like do interesting stuff with their models is

kind of one of the strategies that we are pursuing over here. Very interesting. I mean I loved your last part AI people interested in crypto or blockchain in general. So yes, of course, we are looking forward to understand how AI is supporting this and how AI people can engage more with the blockchain and encourage blockchain at the same time. I remember meeting Sel with the conversation of Gitcoin once upon a time. Gitcoin is a project which has helped supported so many projects in the ecosystem like I can't even remember I think when we started Ethereum Cathode it was initially supported by Gitcoin grant. So that's a great one to begin with. Thank you you both for being a part of the ecosystem sharing about it

and how we reached over here. With the introduction out of the way, let's get into projects overview. Yeah, maybe I can go first on this one. So there's actually two ways that I explain deep funding to people. Let me go with like the less controversial one for now. So the less I will say the controversial part as well later but just to introduce I'll say the less controversial one which is that deep funding basically has three components to it. The first part is what we call a dependency graph. A dependency graph would be like Ethereum is at the center and then we look at all the dependencies of Ethereum. What makes up Ethereum? So things like consensus clients, execution clients, dev tooling clients like solidity, all of those are

like the core repos which make up Ethereum. So level one of the dependency graph is just these core repos and there are about 34 of them. And then we also want to support their dependencies and totally these 34 repos have about 5,000 dependencies. So that's kind of the first part of deep funding which is getting the dependency graph which is not a weighted dependency graph. It's just a dependency graph saying these are the different components. The second part is how do we get the weights to show how much more important one is compared to another. And that's kind of where the machine learning part comes in. Like the total graph has 15,000 edges in it because some dependencies

are used in more than one place. So there are 5,000 dependencies but 15,000 edges in the graph. So machine learning models are trying to predict how much more important one repo is to Ethereum compared to another. So that's where the machine learning part comes in where people are building models to give scores on it. So that's the second part and the third and final part is how do we actually select amongst the various models that are giving weights and that's where we have a jury and Puja is actually one of our jury members who was nominated. So the jury would basically go and give a subset of these 15,000. So they might give 500 ratings of like how much more important one repo is against

another. But the model builders don't know which of the 15,000 we have answers to and which we don't have an answer to. So they have to give a good answer on all 15,000 and we use the 500 that we have got from a jury to select the winning model that then gives to all 15,000. So that's kind of the quick overview of it. That is pretty interesting. Obviously having different components, different components playing independent and important role makes it very much interesting. But I would like to collect your thoughts over here. What sets deep funding apart from any other funding model in the ecosystem other than AI? If you would have to say that this is the controversial part now you that's a perfect segue into the controversial one. So basically a lot of

people have asked how deep funding is different from protocol guild which has a similar mandate of funding kind of ethereum core development. So I think there's basically three ways that deep funding is different. One is that protocol guild does not fund the dependencies of repos. So if something like G has a very important dependency, protocol guild is not funding that dependency right now. It funds the people working on G but not the people who are working on dependencies that G is using. So that's maybe like the first difference that we fund the dependencies of the core repos as well. The second is that we give to organizations and not to individuals. So kind of something which nethermind team actually shared with me

was that protocol guild has helped a lot with retention of their staff that because of protocol guild a lot of their staff don't want to kind of accept jobs from other companies which is actually great retention is one of the hardest issues to solve. So protocol guild helps with retention by giving to individuals whereas deep funding gives to repos like we give money to nethermind for being an execution client or a consensus client and that kind of allows them to do recruiting. So deep funding can let them recruit staff in the future whereas protocol guild helps them retain their staff. So that's the second difference.

And the third and final difference is that um protocol guild is using time waiting. So they basically assign importance based on time that if you've contributed for a long time you have a higher weight whereas deep funding tries to do it based on relative importance that if one repo is more important than another they will get more funding and more money. So these are maybe the three big differences that kind of we fund dependencies of core repos and not just the repos themselves. we fund organizations and not individuals and that the waiting is not based on time but it's based on perceived importance.

Go ahead. Yeah, I just wanted to add like that's you know that's the difference with protocol guild but overall like there are three things that we also identified which you know we have seen in other mechanisms like one is projects only cryb babies get fed which is the projects who actually do marketing are the ones that get funding right and the projects which are really behind the scenes or sort of you know the dependencies they don't really have that visibility hence no funding for them and second is like we have always seen I mean in almost all the grand rounds there is like a group of people you know there are committee members or even for example like community-driven which is QF there are these donors who sort of you know decide who should be

funded or like have a say in it right and then the third is that you need to have opinions on almost all the projects like I still remember when I started at Gitcoin and I was you know we set up GG and I was one of the reviewers and we had 300 projects that we were supposed to review right like it's literally It's insane and so these are the three big things that I see you know deep funding addresses right and somewhere I feel like yeah and obviously like you know Dash also mentioned like deep funding like deep is quite significant which is like dependency graph and also like the deep learning part of it which is where we see AI as an accelerator right so we still have the human mechanism but it's

AI that sort of accelerates it right so uh that's how deep funding sort of addresses is all these three points that we see as challenges in the traditional funding mechanisms right probably humans did not even think of it I don't know because it's just so manual and you know like nobody ever thought that we could actually do this but sort of we see like AI how that you know it brings that unlock yeah do you add anything there Danch yeah so like Puja is a jewer so maybe I'll be curious to hear her experience that kind of the last point Sel said about not having an opinion on every single project something like in a Gitcoin round you have to like have an opinion on every project that do I support them yes or no and like all 600

whereas kind of in deep funding like we focus on quality over quantity that just give us 30 ratings but make them well thought out because that is going to select the model which then gives like 15,000 ratings and our hypothesis is that this is like a better jeweler experience like it's stressful to do like 600 reviews but it's like fun to go into like a few projects and give like a few well thought out detailed reviews but maybe puja you can share about whether that's actually true or not and what the experience has been like as a jura I think I love the concept that we don't have to look at 300 projects at one point of time and the comparison that is there like when we are comparing two project they are generally like on the same platform like whether if that

is a client then other one is also a client and we are trying to figure out what is good and who weighs more there and one more thing that you mentioned is about relative importance. There could be projects which are doing good but there could be a project which may not be doing good at this point of time at present but they have more importance like we want to grow them because they will become future. So this relative importance is also very important. It's not that we are considering the project who is doing at this point of time or have retroactively done better but we are also considering overall life cycle of a project. So I find this concept even very interesting there. All right.

So that's also the kind of advice you've given to jury who've asked that what are the parameters that I should be judging projects on and we basically say it's like s which is specificity. So is it specific to our universe or is it like some library which is used by every single repo. So specificity. The second is additionality which is what you were hinting at Puja which is how well can they actually use the extra funding and can they do something great if they get extra funding. So that's additionality and the third is replaceability that if this repo didn't exist like how easy would it be for you to kind of create it again. So specificity, additionality and replaceability are the three main

parameters we look at. Yeah. Very cool. I know this has been partially covered by SE earlier but curious if you would like to get a little bit deeper into who are the key participants in this project model and what are their roles. Yeah. So I think I can cover the human jury part and then Danch can link it to the AI part. So maybe I can start with the components that are needed for you know to actually run the mechanism or to run the pilot right the most important thing is uh the dependency graph right and huge shout out to open source observer team because of them like we could get the dependency graph right where we have 34 seed nodes and we have 5,000

dependencies and then we have 15,000 edges which is the relationship between them right so again I'll also keep giving TLDDR of deep funding so that people know what it is right so the idea idea is that how do we now assign weights to these you know these dependencies. So what we need is like we just talked about expertise like humans who can do spot checking which means that you know if we can have experts sort of compare these dependencies and then we use that to train the models right so just doubleclicking on the human jury part there are two ways in which the jury has been selected for deep funding pilot one is nomination and the other is invitation right so

invitation is essentially you know we started with vitalic as a jury member and then he nominated two people who he thinks you know should participate as jury members and then they further nominated two people and then this is level one right where they compare the seed nodes and then there is level three which is the comparison between the dependencies themselves for which we need like really technical people devs open source maintainers anybody who is passionate about these repositories or like are familiar with the code are really welcome to be a jury member right because we believe that they have the right expertise to compare these repos.

Right? So this is broadly like so that is how jury is sort of the steering wheel part where you know like vital has also recently called the blog which is AI as the engine humans as the steering wheel because AI needs the direction and sort of we expect human jury members to you know rate these dependencies and give that direction right so that is something that is like a very crucial part of the mechanism and just to also like now that I've talked about the jury members if you are someone who wants to be a jury member please go to research search.allo. And there is a post called join the deep funding jury. You can click on that and then if you think you have that expertise, you have familiarity with the code, you can just drop your name, your email and telegram.

Me or Danch will reach out to you and we will set you up for how you can start making the comparisons. So that's the human jury part and uh I think one important part is that sometimes like jury members feel like conflict of interest but you know that is something that we have also seen like they have the most expertise so good part is that they're not really directly sort of allocating funds they're giving the comparisons which will then be fed into the model right so it's sort of like indirect way of you know selecting models or sort of making the allocation yeah so that is the part that I think one important actor in the mechanism and um then danch if you want to add to it and then talk about the AI part of it.

Yep. I'll just touch on that last part of conflict of interest which you've seen that are a lot of the experts you can't use the knowledge of experts because they have conflicts and everyone wants to work with experts. So by definition they end up having like the most conflicts as well. So that's why like Vitalik talked about this idea of uh indirection that if you're like a US senator and you want to help Elon Musk, there's actually no good way for you to do so directly. At the most you could maybe appoint a friendly FCC commissioner and then that FCC commissioner gives him some contract but like as a US senator I cannot directly help him and that's actually a good feature you know of like kind of to prevent corruption and other things. But

if you look at some of the existing funding mechanisms like say badge holder voting, if I vote for a project, it is as good as me giving them like a $200 bill because my vote is directly equal to money. Whereas in deep funding, when you make a judgment, your vote is only indirectly selecting the AI model. So assume that like I have to give 10 judgments. I'm very fair on nine of them and then on the last 10th judgment I see a project which is run by my friend and because of that I give it much higher marks but because it's only like all of my ratings are being used to select a model. The outlier judgment is not going to be reflected in the model because the

model has certain internal consistencies. So it probably wouldn't like factor in like an outlier to one particular project. So you can actually use the expertise of jurors and just ask them to be as fair as they as possible and like actually use their ratings unlike kind of regular systems where you tell them to not vote but then like some of the kind of more revenue maximizing actors still ignore their conflicts and vote and then the good people get penalized and the bad people benefit even more because they kind of vote. So those are the dynamics we really want to avoid in deep funding. Yeah. So then I can move on to maybe the machine learning part of it. The jury part was very well covered by sage and now we can kind of talk about the other aspect of

it which is kind of in one image if I have to just share with you kind of what deep funding is requiring is basically this is the entire Ethereum dependency graph and there are about 10,000 edges in this graph. So in one sentence we want machine learning builders to give a weight to these lines. You see these lines over here like we basically these are unweighted lines and we require a model which can give a weight to these lines. And once we have that we can put money into one public key and it gets sent across this entire graph based on how much weight these lines have. So that's kind of what we're trying to achieve in like one sentence. And what the jury member is doing is just

basically voting on a subset of these lines and that is used for like some of it is used as training data for the models to be built well and some of it is used as validation which is judging which models are good. We do have the project already live. You can go to crypto pond.xyz XY Z and uh that has the cheap funding challenge and you can basically submit your model there and see how well you stack on the leaderboard. Maybe I can even show you the leaderboard actually here since it's quite fun. I think there's something about leaderboards which just get people's blood flowing that they want to be at the top. So this is the place that machine learning people are submitting.

We've had 5,600 participants so far and uh this is the leaderboard that we have of like people who submitting and this is the score they've got. So the current winner is David Gasquares and kind of it's been changing. Open ledger was the leader last I checked and now it's David. So like it's a constantly changing leaderboard and it's very meritocratic that kind of if you want to get people's attention just submit and if you think you're good as like an as a model builder make a submission and land in the top five you know and then you'll get you noticed. There is another part I want to talk about which is the scoring script that Vitalik has developed but Puja I don't know if there's something else you want to get to or I should deal with that first. Uh no I think that would be great. Please go ahead and

share about that. Yep. kind of it's nice that Vitalic coded up a special scoring script for the deep funding challenge. So if you look at usual machine learning competitions like on Kaggle, they use things like mean squared error or RSME. These are some of the techniques they use and kind of the disadvantage with those techniques is that there is one model which is the winner and that model even if it's good in some parts and bad in other parts it gets to kind of be the winner and like would allocate all of the money. So to avoid that situation, Vitalik made like a script which combines the best of each model. So if Puja's model is really good for judging dependencies of Viper and Sel's model is

really good at judging the kind of like all the level one comparisons between seed nodes. What Vitalic script can do is it takes the kind of puja's model which is good for dependencies of viper sel's model which is good for dependencies of all the consensus clients or all like the all the relative importance of consensus clients and the final leaderboard is that 15% of the mother AI or composite AI is sel and 5% is puja so that's the leaderboard then which is a very neat way of organizing a leaderboard how much are you contributing to the compos Is it AI that distributes money? That's pretty much covered the concept. I was following a tweet by Vitalic like value as a graph

and distilled human judgment. It's a very good combination of both. So thank you so much for sharing about that. I want to touch on a few more points that Sil mentioned in her talk earlier that the dependency graph that was supported by OSO opensource observer. We know that there are a few tools like this in the ecosystem which are trying to help a grant provider as well as grantee to kind of have their accountability listed and shared with the ecosystem. I'm curious to understand if there are other collaborations with such kind of projects such as karma gap and so on so forth or if not are we planning to have that in future? That's interesting. I think one way I think of it is like I

haven't sort of like thought of potential collaborations but maybe there is something there's a there but like for example I do see AI also playing a role in sort of enhancing humans judgment like for example one of the things that I came across when I was speaking to one of the jury members like when they make the comparisons like it also matters like what is available to them like what information is available to them right so I think that is something which I would be interested in exploring in future where can we actually make use of AI in a way which sort of displays information which is more than just what we have right now which is just like very few metrics and that can be more like and for enhanced human judgment right like we do expect

them to have familiarity with the code and everything but Dvanch can probably share we have an example also for one of the repos where you know sort of experts actually rated it very I but when we you know sort of dive deeper into it and spoke to a couple of other projects we realized that this doesn't exist anymore or this doesn't like you know operate anymore so which sort of indicates the need for like AI to also assist humans in making that judgment right which is expertise plus that which is I call it like I think of it as like enhanced human judgment so that is where I see maybe some of these impact tracking tools also could be helpful where they sort of you know yeah help humans I Devanch like what do you think? Yeah, I

think you got all the points covered like karma gap and all of these tools can be used to basically curate the information shown to jurors before they make a judgment which can actually have like a very large effect as well. That is not something we've experimented with so far. We've kind of just let jurors be on their own and making judgments. But I think once the model becomes more developed then we would want to curate that. I think the analogy Vitalik draws is that we want it to be similar to a courtroom. So in a courtroom you have like the prosecutor presenting information. You have the the defendant giving information and then you have a judge who makes a decision. So basically how do we replicate that experience for jurors where we have like good

information synthesized information shown to them from both sides as well. So one of the ideas we had was getting internal and external jurors and seeing if there's consensus. So basically if Puja is making a rating on Viper and then we ask an open-source maintainer of Viper to also give a rating and both of those ratings match then we kind of give it more points and we kind of give it a boost. That's kind of one of the other ideas we're working on. Yeah. Yeah. And I just want to add like if there is anyone who is listening to this because it feels like you know this is like a box that just got opened up. It has so many areas where people can actually dive deeper into. They can like you know do research around it and do more

experiments. So please feel free to reach out and we would love to sort of support you if you know there's anything we can share more on to for you to carry out some research and probably try things with parts of the pilot. That is interesting that you people are still open including some of the things in pilot. I believe pilot project is already ongoing but I was just generally looking into the entire concept and I felt like whatever we are doing right now is based mostly on two things. One is the perception or the knowledge of the juror who is giving and other the information which is already available on the internet. However, we are missing one key point which is called like a

self-explanation providing data point from the project itself in order to say that yes these are our achievements that we count not all the time that can be recorded as an achievement. So I believe that projects like karma gap provides this opportunity to individual project to add that and considering that data point would also help a jurer to make decision. He may or may not be aware of that particular data point that is provided by karma gap. So it's my personal opinion that it would be a good to have this kind of accountability tooling which are already available in the ecosystem be a part of decision making if possible. I think there's like two big issues in the space which deep funding is not solving currently. The first is the halo effect where kind of

that's what projects get you know they can rest on their laurels because they have a good reputation and kind of everyone is going to give them high marks and high donations in rounds and everything like that because of the halo effect. So tools like karma gap are important to actually solve the halo effect that if you look at the last 6 months like this project has not done much so so like don't support them don't give them marks and the second is like a much more niche problem which is kind of what I call a paths out value which is basically like a term from the mechanic and the automobile industry where if you sell your car you actually get less money than if you strip out the car and sell each part individually. So if I sell the individual and tires, I get

more money than selling the whole car. And I think we've seen some of these dynamics play out in like so something like protocol guild gets less money because they are one entity compared to if they had divided amongst many entities and each of which applies in optimism rounds or QF rounds. they actually get more money and like open source observer has done a lot of research showing this that they actually keep like a flat curve which means that actually if a project breaks up into sub projects each of which applies you get more money than if you were just one project which kind of feels unfair. So it's to be seen if deep funding can solve this problem where if you have more projects included in the dependency

graph, do you get more money than if you just kept it as one project? And that's kind of to be seen. Yes, of course. This is just the pilot. Talking about pilot projects, I understand there are about 33 projects for this particular round. What would you consider as like success metrics for this first defunding round? I think success would be that we can like share the weights publicly. I think okay I'll put it this way. It's always hard to give away a large amount of money without controversy. That's like always super hard. So I think if we are able to get weights and give out money to like 5,000 different repos without controversy, I think that would be like a big success. Yeah. And just two things

that came to my mind is maybe this is more on implementation level but I feel like you know one of the things is getting the jury members right like we have been talking to them and hopefully you know through this podcast we are able to reach more core devs you know so the more jury members we have the better it is the more comparisons we will have and second is you know there is a part like Dvanch mentioned in the beginning like crypto people are interested in AI but how do we get AI and ML people interested in crypto so how do we reach to those communities and um you know people who can take part in this competition right so I think that would be and how many submissions do we get how many people are participating I would definitely you know we would I

think count that as a metric that we would want to sort of judge the success of it against yeah I'm sure like there would be more metrics coming up like we will brainstorm there are so many people helping out it's actually collecting human brains as well as AI implementation right here so we will come up with different metrics as well. One issue that I'm actually stuck with in deep funding which like this is where I'm stuck in my thinking right now which is that once the pilot is over and we have like a first draft of the weights that each repo is get is getting what is to stop a model builder from just uh submitting those exact same weights and now they come to the top of the leaderboard. You know they've like

literally just copied it and sent a file and now they're at the top. And if they wanted to be even more like kind of revenue maximizing, they can keep all the weights the exact same except just change the weights to their own project which would then be like way boosted artificially. So that's kind of an even worse issue. So kind of it's a problem of how do we make like these weights have to be public but we also want to keep running this machine learning competition where people are submitting original models but then once you have these weights what is to stop an unoriginal model from just copying those weights submitting and knocking everyone else out of the leaderboard. I don't have a very good answer but I have a hope maybe Vitaliki come up with some different script that may help us. Well,

right now we have those 33 projects, but I was going through one of the posts, right? And I'm thinking from projects which are not a part of this present round. So, I found a post shared by Gitcoin. It was deep funding a visual guide in three easy steps. The article also talks about some projects which are into infrastructure layer and they are currently underfunded in the Ethereum even if they are like kind of critical for tracking so many things network resilience reducing reliance on major cloud providers things like that. I'm not sure how many of the projects are infra projects are part of this pilot round but curious to hear your thoughts on inclusion of such projects in future and how in general you think that the funding will be in a place to help

projects which generally require support and are less visible. Yeah, maybe I can start and then Dan can add. So first of all I just want to say like this deep funding is a pilot right to purely test out the mechanism end to end right and that is why we chose the repos listed in cliniversity.org or right so it was meant to be just this right we had to limit the scope of it the second is that I think who wants to anyone can run a deep funding round and you know can have their dependency graph and right now I feel like it's probably easier to have a open-source software dependency graphs right because through GitHub and you know we have teams who can help with that and maybe a bit challenging to have

a dependency graph which is like not uh open source or not code right but hopefully we will see more of it. So anybody who wants to run or like have a dependency graph, I would highly recommend to get in touch with OSO team and get a dependency graph right and then probably we can happy to talk to them and see how we can actually run a deep funding round for them. So yeah, the idea is to not just limit the you know the projects to whatever is included in the pilot but eventually to sort of expand the graph and you know include other projects and probably run parallel deep funding rounds where you know we have different dependency graphs for different sort of ecosystems who can sort of fund their dependencies right

definitely feels like a powerful mechanism which every ecosystem should be looking into because usually they have grant programs which again go through all the challenges that we talked about where we have projects who are visible they apply and uh probably there is an expectation from the grant giver that they will sort of pass it on to their dependencies which I'm not sure if it's happening and if it's happening I'm not sure to what extent that is correct because it's really hard right to sort of allocate and on what basis do you do that right it again comes down to like a few people deciding probably they have like more knowledge because they know which dependency matters more to them but it feels like it runs into the same issues. I don't even think that is

happening yet. But yeah, so that's what I would say. It's not going to be limited to this and it's just the pilot that we included it. Maybe I can talk quickly about one other pilot that we're doing with GG23 with SAL and her team which is the idea of like predictive funding. So before the round event starts, we are running a competition. And it's going to be live next week actually this like this new competition where you have to predict how much funding a project is going to get in GG23 before the round event starts and then we let the round continue as it is and once the round finishes we'll see which submission best predicted the results in the Gitcoin round and kind of we already ran a mini contest version of

it earlier and people like the winning submission found that number of open issues in GitHub was like a very high indicator of how much funding a project would receive which was like a kind of counterintuitive. So it it combines metrics with actual human allocation like by running these predictive funding rounds and eventually it can even lead to a situation where a project doesn't take part in Gitcoin but then we can know how much money to give them based on what the winning model would predict that they would have got if they had taken part in the round. So I think like applying this idea of distilled human judgment for different juries and like predictive funding rounds I think that's going to be important and the main variable to look at is the error between

the top models prediction and the actual allocation from the jury. If that error rate is very high like we cannot get into this yet and it's like not a good idea to use AI but if the error rate is low it means that we're actually close to replicating distilled human judgments. I'm kind of quite keen to see what this particular data point reveals. This is great. I feel like it should not be viewed as fund providers but it is basically a concept that is being shared out. It's like powerful mechanism that anybody can also copy it or like replicate it in order to fund dependent projects in their ecosystem. So that's very good to know. On the same track I would like to understand here is how

anyone interested with the funding project may be able to contribute as a new contributor. I understand juror are playing an important role in this ecosystem. Sel already talked about how someone can become a juror but people who want to sponsor or grant providers how they can reach out to you and how they can support the funding initiatives. Yeah. So the ways you can support one is as a jury member you can apply on the ALO forum. Second is as a machine learning or an AI model builder. You can like make a submission in the deep funding contest and try to have your model be the allocator of money.

There's also prizes here like Vitalik has committed $50,000 in prizes to people who are building good machine learning models in deep funding. So that's the second way and the third way is by supporting the deep funding pilots. So whatever weights we get basically just getting in touch and we are going to eventually like the main output of deep funding is a splits contract and we're using drips for the kind of distributing money to the repos. So then you can just put money into one public key and it gets sent to all of the repos. So that's kind of the other way that actually once this contract is created just like allocating uh some of the funding to actually be given to these repos through the deep funding kind of weights that get obtained. Very

cool. This has lot of ideas, lots of thoughts around and we are looking forward to have this project stay in the ecosystem as long as possible. Uh we have been seeing Gitcoin around for very long time. They introduced the concept of quadrating funding and other things. So also hoping that the new concept that has been introduced with defunding should stay here for longer and I'll be curious to hear the long-term goal for this project. What's next for defunding? Anything exciting? Any update or goal that you would like to share with this forum? Yeah, I can start and then Danch can share more. So definitely the first goal is to like bring it to the finish

line, right? like there is a lot more work that still needs to go into it and um you know the idea would be to get Judy members and you know get the model submissions long-term I see more folks of deep funding like so like more ecosystems running deep funding rounds right funding their dependencies I also hope to see which I'm already seeing like you know there are predictive funding challenges that the munch is running and um there's deep which is another interesting project don't want to talk too much about it like still you know in the making but shout out to David Dao from gain forest who is sort of leading that it's a folk of deep funding and sort of you know few elements changed into deep funding and like totally new way of doing it so

hoping to see more you know ways or versions of deep funding that are being run and I hope to see that you know deep funding relies on this basic underlying principle which is AI as the engine and humans as a steering wheel right so deep funding is sort sort of an use case of this so what I hope and I hope that I can also play a part in it like creating more use cases which has this underlying principle where you know you have the human mechanism and you are using AI to sort of accelerate that right so I think this goes both ways you look at the mechanisms that currently exist and if those can be improved in some way right and then what are the new mechanisms or what are the new ways of funding where you can use this basic principle and

sort of you know do that so that is what I hope from you know deep funding and long-term one thing that I personally think I would be interested in long-term to sort of you know devote my time to like as Vitalik talks about like public goods funding has these two problems one is like where does the funding come from and how do you allocate so you know we are really focused more on how do we allocate right so hopefully we also transition to and like work towards like where does the money or capital come from right like how some of the projects can actually generate returns while still pursuing you know a social cause or like a work which sort of you know is a public good is something that I hope personally to be get involved in

eventually and this space also sees more solutions for yeah yeah I think like the main thing which I'm excited to see is like that we have so many partners who've made deep funding possible like we've used pairwise for some of the voting comparisons we've used drips for allocating money pawn to host the competition oso to generate the graph allow to kind of bring people together and the full website and other things that they've done. So kind of the other part is just how we create like some form of like a protodow which anyone who wants to run a deep funding instance can approach the protodow and get their help to do it and how we actually make it sustainable for everybody who's been involved. Right now kind of a lot of

people have been doing it because it's a new mechanism. A lot of the partners have been contributing on that basis. But how do we actually make it sustainable for the work that they've put into deep funding is kind of something else that I'm excited about. Very cool. In this series, we have covered different kind of public good funding concept, public good funding providers. People who are new to this uh series may also watch the opensource observer episode. We did karmaga, we did octant, now we are learning about deep funding. So we do have a lot of mechanism in place to support different projects coming up in the ecosystem. I'm sure people who are getting involved they will help us make this system even

more evolved. So I'm hoping that the answer to Sel's question she would be able to in a position to help where does this money come from? There are so many people which I think they call it dragon and legions right in Octton term. So they would be around to support us and they would be in a place to help us get this money in place and with the help of deep funding we would be in a position to distribute it to the right and project that actually deserves the money. All the very best to the team and I look forward to have this team getting success in their pilot round and maybe have many more rounds. I don't know what is the cadence that you are looking for quarterly, yearly or whatever it is but

best wishes for all of them. Thank you you both for joining us here today and sharing about the funding. I don't know if there is an opportunity we can perhaps become the community partner to promote about defunding. You have been collaborating with so many different projects. We can start talking about defunding and reach out to community to participate more support more and things like that. Yeah. Thanks for bringing us on Van like this was a great podcast. Yeah, thank you. Sure. Thank you so much for joining us today. So that brings us to the end of today's episode. Thank you both Sil and Dvanch for sharing these valuable insight. If you want to learn more about deep funding, check out

website and social link shared in description. I hope people may have learned and get inspired. be a part of this project and any other project of the ecosystem where they think that they can add their value. So I love to hear your feedback on those. If you find this episode interesting, please feel free to share it with your fellow contributors, fellow members of the ecosystem and stay tuned with Ethereum Cat Hurdles. Thank you all for joining us here today.