Shorttalks Metagov
Speaker 1: At any time when that there we go.
Top Keywords
- reputation 0.009
- token 0.006
- research 0.006
- tensorflow 0.005
- learning 0.004
- time 0.004
- open 0.004
- reputation systems 0.004
- daos 0.004
- governance 0.004
- token voting 0.004
- pytorch 0.004
Transcript
At any time when that there we go.
Oh, got it.
Perfect time. So it is now recording. Yeah. So as I was saying, our our whole goal at Sprup is really connecting industry and academia. We're trying to maximize for, how we can distribute and create incentives, how we can create access and facilitation between industry and academia, how do we make more information from academia accessible to folks in research. And our real goal is to become a place that is known for being a well moderated and run community and forum as the starting point where we help produce quality interactions, where various kind of research minded content comes out, and where we, as an infrastructure layer, try to provide support to different groups in various capacities. So just to give you all a sense, we are actually already working with a few folks in the meta gov community. We're support supporting a gov based researcher. So with some of the work that that Anne Brody and Josh are doing there, we're supporting two of Nathan's projects in terms of exit to community and community rule, and we're actually working with supporting Seth in terms of a literature review that he's working on with some students of his. And so we're we're looking to expand that and work with more folks. So definitely one kind of a small call to action is if you are working on any research, whether it's formally through meta gov or an academic institution or just individually, but you're actually looking for some kind of support, please let us know. And, again, part of the mission is for us really trying to run a series of experiments for, you know, how can we provide the incentives to get more of that interaction and facilitation going so that Web3 can move faster and learn faster and and just mature at a quicker pace. I'm sure for anyone who's been, you know, in governance discussions for five, six years, or at least if you were, you know, pre the 2017 boom, I'm sure you might have been in conversations where you're like, oh, a lot of this DAO stuff sounds like co ops, and people from the web three space would be like, oh, it's a co op. So now, you know, that has matured around governance. So we wanna help make sure that that kind of maturity and learning is happening everywhere that it could, and our starting point is information availability. So this is an example of Patrick Zeus as a grad student out of ETH Zurich. I'm forgetting which European, I believe, Swiss institution specifically. But, anyway, he was finishing his grad work. He wrote this paper on analyzing and preventing sandwich tax on Ethereum. So we were able to give Patrick a grant so he could write about it on our forum, get some additional engagement and interaction going around it. And that's exactly the kind of grants that we wanna be extending as well. So if you're also if you have already produced output and you wanna make it more accessible to folks in industry or maybe you want to collaborate with us on the podcast that we're going to be launching later this month, which will actually feature a few MediGo community folks as well. You know, we're starting to plan future mini series as well around the podcast. So there's definitely a lot of fun activities, and I don't wanna bore y'all with all of these details, but these are just some of the various areas of of kind of activity around what we're doing. You know, the actual content pipeline and producing content for the forum is our main activity to start. But a big one that we're leaning into right now around review is we're trying to understand what does review mean internally for the kind of summaries that are being produced. But one thing that I'm very excited for us to get to work on in the coming months is what does open peer review look like in decentralized research environments? Because there are now a number of DAOs that are and I'll share a link later, but I'm putting together a list of research DAOs. And, right, most of these DAOs, possibly with the exception of MetaGov DAO, because at least this one is run by academics. So I I have less concerns about some of the peer review aspects, but a number of these research DAOs have no one from formal research background. So what is the quality assurance layer? What does peer review actually look like in these environments? And I'm not gonna bore y'all running through this whole thing, but I I just wanna say that whether you're interested in getting grants to do something more research oriented, if you're interested in helping us figure out what peer review could look like more broadly, or if you're interested in project management, please let me know. We have a lot of opportunities internally, and I'll specifically say we have a governance specific project management role that we're creating right now, which is someone who could help with things ranging from grants, strategic partnership, and club outreach as well as events from an outreach perspective, as well as the kind of information that's hitting our pipeline and the podcast that we're launching and just keeping all the things coordinated there. So if you're interested in any of that, let me know. But, yeah, the ways we generally collaborate are extending grants. We have a lot of opportunities if anyone is looking for part time work. We're currently doing a DAO day either in early Feb, but we're doing a bunch of events now as part of ETHDenver. So we might push that out to March, but we're going to be at ETHDenver. We're going to do a four hour event on what does decentralization of research mean. We'd love to collaborate with anyone there who's going to be there, and we are most likely going to be an official sponsor of ETHDenver. I will be circling back with the MetaDev community because we will then get to plan a nontechnical, non code bounty, and we wanna do at least one strand of that around governance. So, yeah, this is kind of us, and I went a little over half the time there, but I wanna leave at least a couple minutes for anyone who who might have any questions or anything like that. And, yeah, so thanks for dropping with Daniel Ospina's Research DAO. So, yeah, that is definitely one that is on my list. And let me pull up my list so I can just share that. But, yeah, Daniels, Toby Shorin, and other Internet are here though. Then MetaGovDAO, Governance is another project, but, yeah, I will drop a link. So, yeah, definitely happy to to chat with anyone. I just dropped the link to that spreadsheet that I created, and that's proper research DAOs, please let me know. As mentioned, we're gonna have this event in February talking about the decentralization of research and what that means and how it can all go horribly wrong. I might ask some meta gov DAO to prerecord something because I know Kelsey is not planning on coming to Denver. And then we might actually get Vita DAO, open access or not open access DAO, a decentralized science DAO and one or two other ones that I'm forgetting off the top of my head. But, yeah, I I'm really excited to start having more of these what is decentralizing research peer review and publication conversations more publicly. So, yeah, happy to answer questions. Otherwise, happy to cede the remaining time to keep lightning talks going.
I was literally just gonna ask for this web treat, and thank you so much for sharing. I'll try to add to it because I get a lot of outreach from folks like this as well.
Awesome. Yeah. And please, yeah, add to it, share to it. I'll just add you as an editor right on that, Josh. If anyone else wants to actually ever add to it or just let me know names, just please feel free to drop me Slack or Discord or whatever other Messenger.
App. You maybe just have a pointer. Like, this is super interesting. I'm also really interested in decentralized research. You like a pointer to any, like, particularly, I don't know, like, tooling that you think, like, or blockchain effort or I don't know. Like, just something to start with even. I mean, I see the names of all those different DAOs, but just like a good primer.
I'm muted. I'll share a link to the specific to the specific summary that I was highlighting as well for the the Patrick Zeus sandwich attacks one. So, yeah, we're always excited for folks who just might want to, come in and potentially write a research summary and get a grant for that. I just dropped the link to our idea proposal form. You can just fill that out and our team will follow-up with you and you can get, depending if you're a primary author or not, you can get anywhere from a couple $100 grant. If you're a primary author, you can get in low thousands. But, yeah, we're just our whole goal is to try to think of if if there was a group that has money with the mission of how do we fill the gaps of research infrastructure, we're in a fortunate enough position that we are in we are there. So how do we fill those gaps? So we need to be aware of the gaps as much as possible. So as much as anyone here wants to let us know, like, it would be great if someone covered this area or facilitated these conversations or just built a database and paid for people to manage it, we're very open to any and all of those things. So, yeah, please feel free to to look into our forum and our resources. I'm happy to send you share more links, but I see I'm hitting time, so I'm gonna stop talking to keep us on track.
Thank you for being such a great citizen of lightning talks. This is wonderful. Yeah. I think the intent of these is to seed ideas and definitely continue conversations offline, and so very excited to see that already happening. We'll now turn it over to Ronan who was already speaking, amazing, for your lightning talk.
Hi. Yeah. Thanks. Let me just share my screen.
I just have some slides. I don't know, like, how if they fit in ten minutes, but I'll try to sort of see where the interest goes and try to do some time also to talk. Yeah. Cool. So this is my first time presenting anything in MediGov. So thanks for the opportunity, and, yeah, looking forward to engaging with you. So I am working at Common, also called DAOstack. So DAOstack has been a long time in, like, the DAO arena, and common is kind of a new outgrowth of that, trying to connect it more with, like, public goods and concrete, like, impact questions. So this presentation is about something called common sense, which is a human centered sense making data over the web. It's more of proposals. It's things we're trying to think out now. And, yeah, let's dive into it. So, yeah, the motivation here is kind of coming from this information overload. I think, I mean, most of you are familiar with the kind of, this kind of problem. You know, we have this ever growing exponential wealth of information, and this creates a poverty of attention. And, yeah, we need to figure out how to allocate our attention in a better way. And this is, like, a huge problem, of course, and it's kind of getting more and more, I guess, you could say urgent. And it's you know, anywhere you look, you can kinda find where things are going wrong really fast. This is just a recent, like, talk of Daniel Schmachtenberger and Tristan Harris. And, yeah, it's like, how can you run a democracy if people don't have a shared sense of what's going on in the world? And it's also in academia. There have been, like, some recent papers on this kind of idea of, you know, how do we steward global collective behavior? People are saying that it should be a crisis discipline just like medicine, conservation, and climate science have become. But there's something that's, kinda stopping that, which is that most of our information foraging processes are are, like, offloaded onto the algorithms, and, you know, we all know they're they're not, aligned with, like, incentives for, informed and and and just society. So because they're proprietary in Black Box, they slowed down all these efforts for actually trying to learn how to, you know, do this kind of stewardship of of global collective behavior. So this is yeah. It's a it's a big problem. I mean, I I kinda just got into this, and I've been sucked into it. I'm like, why am I working on my PhD? Like, this is so much bigger than what I was doing earlier. And the question is, how do we build a healthier information gathering environment? And that's kind of a huge question. So we try to break it down to something smaller. How to govern our individual collective attention that's even also pretty big. But this tweet yesterday kind of helped me put it more into words. I already had, like, 100 and some thousand likes just for one day. So, yeah, grant me the serenity to close all the tabs. I'm never gonna read the courage to read the open tabs. It really ought to be better. The wisdom to know the difference. And, yeah, this is what we want. Like, this is a concrete problem we wanna help solve. Like, how do you do this? Like, all of us have, like, millions of tabs open,
and we don't know what
to read and what not to read. And Internet isn't really helping us there. So, like, inspiration, we're we're kinda going back to, you know, cognitive science, sort of how children learn to navigate this kind of world, you know, the boobooing, buzzing, confusion of of of early childhood. And there's this idea of shared attention where you have, you know, teachers or parents, authority figures that that show you, like, what to attend to.
And it
it's like this crucial thing. It's kinda seems almost obvious that when you think about it, it actually doesn't really exist on the web. So so there's, like, this gaping hole and maybe causing a lot of data problems we're seeing. So this is, Zach Sine. He's, like, a future of education thinker. I highly recommend. And he says, yeah, if you put a kid in front of YouTube, YouTube is just gonna capture the kid. And the kid is gonna be seeing a sequence of videos just, you know, because this incentive to capture attention. But, actually, there's a lot of really good sequences the kid could be watching, like, for some historical topic or whatever topic he want. They exist, but they need to be curated and not by algorithms that are mining attention for profit. So, yeah, this idea of, okay, we need to sort of curate these web trails. We don't expect to be able to, you know, navigate these mountains without good maps, and can we do the same for the web? So that's kind of the motivation. Yeah. The idea here is I'm really going to this mountain metaphor. Can we overlay web content with, like, this kind of semantic tags, like handles or web pages or or or other web content, and then, you know, help people navigate even more so and navigate using that. Yeah. And there are a lot of things that kinda sound like this. Hypothesis, Memex, tons of of cool things in the space. I think the thing I wanna focus on is is that, like, the Web three, this is kind of the dawning on me that, like, we should be thinking about protocols, not platforms. A lot of people have said this already are just kind of echoing it. But I think that addressing the, you know, digital tension crisis is way beyond one platform. And I kinda think of it like, you know, this, meme. You have all these little platforms. I mean, some of them are pretty big, but they're still little compared to TikTok and Facebook. And they're just not gonna cut it unless we can somehow organize, all of these. And then what we need to organize, I think that's a, you know, big question we're kind of going into, but there should be, like, a protocol that facilitate interoperability in open markets between all the different components that such a that such a sense making thing needs. And, you know, there's data storage and recommendation, front ends, all this stuff. How much time do I have? I kinda missed when I started. It's, like, six minutes? Or
Yeah. You have five more.
Five more minutes. Cool. Yeah. Cool. So, yeah, again, just like I I don't I'll skip over this because it's kinda long, but the idea of protocols and our platforms is that once there's, like, a common protocol, then you would have third parties creating whatever interfaces you wanted, whatever rules, and it wouldn't fall in, like, this one bias centralized platform. Yeah. And then the idea is, like, empower the ends of the network, the communities, the public industry organizations, competing platforms, whatever. So what are we suggesting? Like, more concrete? Like, I think we're trying to figure out what is a core interoperable object that we make, sort of constitute this kind of protocol. This is the current idea we have. I'm sure it's gonna change more because we just got started with it, but we're thinking of this like web hold. You know, going back to the climbing metaphor. So this kind of hand hold is, thing that can has you place it on a piece of text or on a span of text or sorry, web web page or a span of text. And, yeah, these are the kind of basic affordances. You can tag it. You can link it to other web holes. You can comment on it, like it, and set visibility. So these are, like, the basic functions we we see right now, and there's a lot of applications you can develop over something like this. Before that, I guess oh, yeah. One thing about the, you know, back end. So that's kind of that's kind of you could think of it as, like, the shared attention because we're putting the like, these web holds are are in some shared space, which is, you know, should be, like, have permission to privacy control and decentralized some kind of decentralization as well as open access. So, yeah, this is kind of in forefront of, you know, the Web three d storage technologies. And then, you know, you have some interesting potential uses where you might have users that mark content, like, you know, writing to this. So that's existing sort of social bookmarking or web trail mapping. Apps can sort of we would like them to interface and, you know, pour their information if people want it, of course. You know, they would that's obviously all things that people opt in and, add to the database that they want. But people do it anyway on Twitter. Right? So people do like to share knowledge. So they would be adding this kind of information, then you could have all kinds of interesting, like, read sort of interfaces. You know, you could imagine you're, like, navigating on Wikipedia and you see this up there and you kinda see, you know, oh, there's some, you know, interesting rabbit holes here. I can look at these playlists and and, you know, join them, meet people, meet content, all that stuff. There's another app called AmpEmpty, which kind of already does this in a limited way. It kinda takes some Twitter and y combinator threads, and it backlinks them to different content. So that's also a cool way to show, like, how this stuff can really help, navigate the web more effectively. But, again, we'd like to pour all that into the same common database. Yeah. And I guess in the future, I don't think I have time to go into this. But, like, Fitz eventually, we'd like to do some kind of, you know, fixing search and recommendation because, you know, the database is is open, so you can have, again, the competing kind of interfaces and and and recommendation systems. I would like to leave time to talk, so we've got just kind of cut to the yeah. Yeah. Key questions are, like, how to not become a platform? Is the full Web three tech stack ready? How to do this kind of coalescence? And, yeah, that's Daueck or Discover or any acronym that Daueck fits the fits better on the syllabus. So, yeah, thanks. I don't know how
much time we have for questions. Maybe that
was longer than I expected.
No. We have a couple of minutes for questions. I wonder if it helps to ask. Share your thoughts.
Feel free, like, also, yeah, any kind of constructive criticism you didn't understand, but it was too fast, high level, low level. I don't know.
Josh asks, what's the HAO? Oh, of course. Go ahead.
Oh, I saw it just today. It's like a human autonomous organization. It's like trying not to be too dis I don't know. People are just coming up with different acronyms that fit their taste.
Mhmm. I was curious
to hear you talk a
little bit more about search. So I've I've been coming across, like, this idea more and more that, like, search is actually, like, really, like, relevant, like, within, like, a localized community oriented context. But, like, in a global sense, it's actually a lot less useful. I was wondering, like, what, like, the different kinds of scopes of search are that you're imagining in that kind of third use space.
Yeah. Yeah. I think that's a great question. I I feel like, yeah, people a lot of people are saying that, you know, Google has kind of lost its sort of yeah, it's kind of losing its edge in some sense. Like, it's becoming less relevant maybe. And like you say, maybe people are interested in sort of hearing what the local yeah. More of the local, like, what people around them are attending to or or or thinking about. So in some sense, it's becoming more a question of curation of, like, localized. Like, we call it, like, personalized page rank. You could think of it. So if you have this kind of more local information about what your peers are are looking at. So that's one, yeah, that's kind of one one way we're we're looking at. I don't know if that answers your question. But, yeah, this kind of database seems like it's kinda necessary if you wanna build some kind of more
personalized search. Alright. Thanks.
Briefly, if you have one minute, Anna asks what if the issue wasn't attention but literacy? Just if you have thoughts on that.
Oh, yeah. Sure. Yeah. There's lots, yeah, important probably, like, thinking, like, how this could go wrong or how this isn't, like, what we need. There are a few problems here. Right? Like, the information ecosystem, like, there's many there are many problems, and I think one of them is this problem of, yeah, I was trying to make it concrete by, like, the you know, you have millions of tabs open, what to look at. So that's one of the questions we're trying to address. And there's also question of literacy, but you could imagine sort of I think this is this is where, like, Zach Stein is going with this kind of, ideas that you have, like, some kind of curated web sequences, that would also help literacy because they start you off from, like, easier things and then build up your literacy to, like, harder stuff.
Great. I think there are a couple other questions.
I see. What
is the AMP?
Oh, I'm sorry. Probably I'll have to switch over to Jack, but feel free to answer them in the chat itself as we go. Sure. Thanks. Jack, floor is all yours.
Hey. How's it going? So I'm Jack. My background is in, like, startups, data science stuff. I used to do, like, analytics for these for for just for companies just doing business intelligence, and I switched over to working for this DAO called Index Co op.
Let me share my screen. I don't have permissions.
I might have to exit and come back in. So because I don't have permissions here. Alright. Be right back.
Ronan, if you wanna take a second to answer the other questions, you can go.
Yeah. Sure. I just said I I put in the in the chat the ambi app that does, like, the Twitter backlinking. And then you said, could personalized page ranking amplify lack of shared information ecosystem? Sort of being like echo chambers, so you're saying. Yeah. Yeah. Yeah. Yeah. Definitely. I think also, though, that, like, opening like, having this marketplace of of of different interfaces would sort of allow you to help with this kind of bubble filter bubble because you would sort of you know, you're you make it more like, you know you're choosing a certain kind of search algorithm or recommendation algorithm, and then you'd know from your peers, hopefully, that there are other kinds, and maybe those ones are doing something that's, like, trying to burst filter levels. So yeah.
Great. Jack, great to see you back. Hope the permissions are wonderful.
Alright. Okay. So I end up working for this decentralized autonomous corporation, essentially, where there's a governance token token voting. Right? I'm working on their operations. They they have cash flow. They're like a whole you know, basically, like a startup without a head, you know, typical Dao stuff. I started getting pretty involved in their operations. And, essentially, like, as I got more and more involved in, like, how this stuff works, I started with wanting to study, how, like, the, like, what are what practically speaking, how does token voting actually happen on chain or, you know, whatever? And, like, my skill set is in, you know, data analysis and stuff like that. So my inclination was basically just to study it myself. And I ended up, writing these, these posts essentially, that sort of studied, like, what happened in terms of token distribution, how what what looked like, what what the actual historical governance process was. And I have all this, you know, open source code now to pull snapshot data because we do all of our governance on snapshot. And what I'm looking for in the code, generally speaking, are critical voters. Right? And not from, like, a beds off power index, like, you know, abstractly, you hold a lot of tokens sort of critical voters, but people who are practically dominating these these proposals. Right? Because the proposal rate is so low historically in Dallas that it matters. Like, you can have a very, very small percentage of the total token distribution, but still become the single deciding voter and actually single handedly to make these decisions. So I kinda, like, talked about it with these guys. There are some things that they are working on to kind of mitigate this, but, functionally, because these guys are interested primarily in just making money and growing the DAO, being on the forefront of governance is not necessarily their interest. And, essentially, what I'm looking for here, when presenting this research and, like, my findings to you guys and kinda try to connect to you guys is looking for additional, like, collaborators, some feedback, some understanding of, like, how does this fit how does the stuff that I'm doing based off of, like, totally different direction? How does that fit into what you guys have been researching and what you guys have found regarding, like, token voting? Is there any appetite for expanding this kind of research to, like, pretty much every other snapshot? You know, every like, the the the the code that I wrote works for all snapshot proposals, and you can expand the research to everybody, to Sushi, Uniswap, Gitcoin, Olympus. Right? But I don't have, like, any sort of institutional, reason to go into their system and, like, be like, hey. What's going on, guys? Like, I just analyzed your your your, your governance system. Whereas I had that sort of, like, backing for for index. So, I just wanted to get your thoughts, reactions, yeah, essentially.
I'll just quickly jump in and say that I I have not had a chance to read this post yet, and I'm very excited to. But, overall, this is very much in line with the kind of stuff that we would love to support and and help create more of, and we're really interested in doing more nuanced case studies in different protocols and actually getting down to granular components of, you know, snapshot governance voting on snapshot for specific ecosystems. And I'll just say that we might be open to potentially funding some elements of this as well as trying to help and build some of the connections and and bring folks together and whatnot around it to to get a bigger crew. So, yeah, I'll just throw that out there.
Sorry. I I couldn't see. Who was talking just now?
That was Eugene from from Smart Contract Research for opting you directly.
Alright. Great.
Something I wanted to ask. So what's the like, suppose we find these critical voters. Like, what is that what's the next step after that? Like, what is the like, what is Index, you know, hoping to do with that information?
So, I mean, that's like a whole that's like a separate there's like a, like, a data analysis section that, you know, I was doing, and then there's, like, theory section of what we're what we what we can do about it. And, my proposals were essentially to create, like, a, like, a board, essentially, of activists, like, you know, critical voters, and basically make them more transparent. And the other idea was to actually take decision making out of token voting. Essentially, basically, conceding that given the mechanics of how tokens are being distributed, there's going to be, like, a significant concentration of, like, you know, voting power there. That that was my proposal, but there's there's some, like, there's some conversation around it, but I wasn't really I didn't really have as robust of a, like, a discussion or conversation around the pros and cons of that as I as I thought. And practically speaking, Index Cop is taking steps on both of these. So there is, like, an investor relations, like, you know, thing that anyone can join, And there's also an increasing push towards, contributor led decision making that doesn't rely on token butter. So both of these are practically happening.
Interesting. So okay. So you want to get rid of token voting in index, but what you're analyzing ultimately is still token voting. And the point of the analysis is to illustrate how kinda skewed it is in some sense.
It's not that I wanna get rid of it. I I think I mean, like, it it serves a very important purpose, right, which is, like, it the the legitimacy function of token voting is extremely important, but it's too slow, and most token voters most token holders do not have the necessary context. Right? The the basic divide in index call, but I think is actually probably true broadly across a lot of doubts, is that I mean, this is probably true capitalism in general. Right? The token owners are capital owners and then the people who are doing the work and the tributors are the laborers, and the people who are doing the labor have all the context. Right? They they they have a lot of time and they contribute and they have and they they can make really high quality decisions, but they don't have the equity to reach Quorum by themselves. Right? That's the basic dynamic within index coop itself. Mhmm. So that basically, that tension is kind of what I was highlighting. And I was like, this is not necessarily that token voting is illegitimate, but it's just that the token the people who own the token just don't, don't have the context. And then a lot of people so, like, basically, my two proposals were give people who have context more power even if it's not democratic, and then give, people who have a lot of power more context. Those are, like, the two ideas.
I'm curious about how you've seen, I guess, I'd love to hear examples of where you think the the worker space kind of has a higher quality decision, but the coop went in an or sorry, index went in another direction or kind of because one of the interesting things about worker coops is often that, you know, the idea is to align incentives between the workers who want the company to grow because they themselves have a stake in it, etcetera. And this often works really well, but there are some downsides. One being that, you know, workers and worker co ops often are not into hiring very much. And don't wanna grow in certain ways because hiring more people, even though it may make it more productive, may lower wages. So there are interesting incentive systems at play in worker cooperatives that I think folks are are constantly researching. So I'm just curious as to, how that transfers if at all to the situation you're describing.
A lot, actually. I think when like, a lot of proposals have been kicked around. Like, essentially, the contributors noticed almost immediately. Right? Index co op is maybe about a year old, and contributors noticed very quickly that we didn't have any legislative power. And there are a lot of different proposals that were basically made. They're like, hey. Why don't why don't we turn this into, like, a democracy? Why don't we figure out how to do proof of humanity? These ideas all came up, but I think they end up being rejected primarily because I think it was just too too far. Right? In the sense that, like, when if you talk about a pure worker cooperative, I think a lot of these downsides make a lot of sense. But since we're starting the the starting point of the next call is, like, a pure equity, like, owner owned situation, the any movement you go towards a worker cooperative is likely not going to overshoot the point at which you are now, like, incentive misaligned again. So, like, it's sort of like a directional, like, people can sort of sense where the inefficiencies are coming from and they're pushing towards that inefficiency. And there are there's also because it's so decentralized and because the token motors ultimately control everything, they're less inclined to be like, alright. We're giving up all of our power and now it's a pure democracy of all your contributors. Right? That that didn't happen. It's not likely to happen. So to answer your question, I think people are aware of that, and I think that's why we're kinda moving steps towards a worker cooperative without actually fully becoming a worker cooperative.
I this is super fascinating, and I there are other questions in the chat. We'll I'll switch over to Max now in the interest of time, but hope we can continue learning about this.
Well, hopefully, I can screen share. Okay. Looks like I
think I have host permissions, but, Shauna okay. Great.
There we go. K. Let's try. Oops. Can everyone see this?
Yep.
Cool. Okay. Hello, everyone. I'm Max, a master's student at MIT, and kind of I work on stuff with Josh and and and MediGov. This isn't a crypto related presentation, but it does have to do with governance. And so, you know, I'm pretty glad to be here today. I'm gonna talk pretty quickly about some of the context behind my research methodology that I've been employing some preliminary findings, and I'd love to hear some feedback. So so, yeah, my research is trying to understand, the dynamics, specifically the institutional dynamics behind the open source machine learning ecosystem. And so for some context, you all may or may not know that, basically, all research in in machine learning now or deep learning specifically is done with an open source, framework like PyTorch or TensorFlow or Julia. And, this is like, this broader thing, I think, is really important because, you know, deep learning affects, like, many, many, many people. And in the future, if you think that the impact of of or if you think that machine learning is gonna continue to grow as it has been, then I think there's a very strong case to be made for the fact that these tools are going to shape how it's going to be implemented in the future. And, so far, there's been, you know, very little examination of this the institutional dynamics that, that that shape the tools. So and and the institutional dynamics, I think, I I equate with the governance here too. And one one other reason I think it's especially interesting to look at governance now, of the open source machine learning frameworks is is that around 2016, there was a really big shift. I would say if there were, you know, two main points in kind of open source machine learning software, 2009 was a time when there was no open source machine learning software. I mean, there's a funny John LeCun paper where he talks about people needing more machine learning frameworks. And 2016, so between 2009 and 2016, there are all these, academic research labs building frameworks like CAFE and Diana. And 2016 was the the era of the shift from the, like, academic deep learning frameworks to institutional or, like, big corporation deep learning frameworks. Amazon, Google, Facebook, Microsoft all have their own versions of frameworks. And so, this this shift that's happened has also, you know, led to a lot of, other changes. And so, you know, as I mentioned before, the three kind of biggest deep learning frameworks now are TensorFlow, PyTorch, and JAX, and they're respectively they're either built by Google or Facebook. Yeah. I mean, there's there's there's some others that are important too, but those are the those are the main players. So I'd say that the main research question that I have here is how do institutions, specifically these big institutions like Facebook and Google, how do they shape the development and governance, of open source machine learning? And how I've tried to tackle this is, is to take a lens that's inspired by, Elinor Ostrom, kind of like an institutional analysis and development lens. And I have first looked at or I'm first looking at different key moments, from the perspective of the institutions. And then in and that that involves, like, looking at a lot of the documents, looking at looking at, the release notes and how things have changed over time, and supplementing that with some qualitative interviews, with a number of the of the key developers. And, ultimately, I think that the some of the most illuminating points to kind of I I was I'm borrowing somewhat from Gabriela Coleman, the anthropologist who looked at hackers, is to focus on what she calls, like, punctuated crises, and and to hopefully use those to look at the dynamics of how, this ecosystem has evolved over time. So why is this changing? Okay. Cool. So the first actually, since I have how many do I have, like, three minutes left, Divya?
You have six or sorry. Five minutes, actually.
Okay. Cool. Cool. So, briefly, from an institutional lens, here's, some highlights. In 2016, Google released TensorFlow, as I as I mentioned before. That kind of was really seminal in the history of open source machine learning. And they were mainly trying to get they were they were trying to build something that was flexible enough both for people deploying machine learning systems and for people doing research. And, you know, previously, they had a kind of relatively closed source framework called DIS Belief that was supposed to be for massively trading things, but that that failed. And apologies if this sentence isn't finished. I I these slides are not quite as polished as I would like. But after TensorFlow was released, it was really unusual in that they were open core. You know, previously, they had had, like, their their page rank. Like, a lot of the other infrastructure hadn't been open source, and so I think it was a it was a big moment for the community. And then PyTorch, came out of, like, Facebook AI research, just a couple of individuals there, and it was targeted much more specifically at researchers who were trying to build flexible models. And the community was initially very excited about TensorFlow, and it's really interesting to see, that they turned rapidly. And then and, like, among the developers that I've talked to, there's much more of a favoring of PyTorch because it's not seen as part of this, like, monetization, scheme. So Google also has their 10 their TPU, their tensor processing units, and their whole kind of cloud infrastructure. And a lot of people see TensorFlow as, like, an attempt to bring people into their ecosystem and to and to monetize that, whereas PyTorch doesn't have the equivalent there. And here, I'm also there's also I'm mentioning ONNX because Microsoft and at around 2018 or 2019, at the level of what's called, like, intermediate representation sorry about this noise. Between the frameworks themselves and the hardware, there's there's, how the models are represented, to be to be shown to the different pieces of hardware. I'm not explaining this very well, but, basically, it's a lower level of the stack. And what was interesting is in 2019, Facebook and Microsoft did a kind of a sponsored sponsored this effort to build an intermediate representation that would effectively allow people to break out of the TensorFlow ecosystem. And so I kind of explained what I was going to talk about on this slide, but I think there are two really natural interesting points of tension here, which is this, like, tension between the openness that PyTorch and ONNX is trying to play for. Like, people are trying to escape the Google TensorFlow ecosystem and, the initial difficulties of of TensorFlow in in bringing people in with various community support. Yeah. I'm I'm rushing through this a bit because I still I would like feedback from people. But, you know, broadly, I'd say the preliminary learning points here are that one is deep learning frameworks have become an arena competition, But where the game is different, like, the different institutions have different incentives. TensorFlow ostensibly does want people using their cloud compute, whereas, like, PyTorch is a little more complicated. It seems to be more motivated by research interests, and that there are certain kind of predictable dynamics. Like, I didn't get to talk about this much, but there is this movement more towards down the stack. Companies are now competing at the, like, intermediate representation level. Yeah. And and I think this, again, is is pretty it's interesting, but also useful because where the attention of these big corporations are, these big institutions is going to dictate a lot about the dynamics that that shape how machine learning is deployed.
Max, just to interject, if you did want to get comments, we have, like, one and a half minutes left.
Okay. Yeah. I'd like to get comments. So, yeah, generally, comments would appreciate thoughts on the methodology, any questions or angles in particular that were interesting. Yeah. Thank you for being patient with this. And, also, thank you to Divya and Seth for for giving feedback on prior versions.
Great. I've got one for this one if, if there's not a queue. Is there a queue? I'm sorry, Divya. Actually, there's
probably Go
ahead. Well, so, you know, focusing on these kinda lower level, platforms and these more production focused platforms, it makes me wonder if, you know, the kinda great thing that Ostreammy stuff's good for is, like, here's some random people, with no resources who kinda have to come together for their livelihood. I wonder, you know, this is almost more of a b to b product, a lot of these, these particular open source projects you're looking at. And I wonder if they evoke more of a corporate partnership, you know, dimension of self governance, which is, you know, which is interesting. There's there's a lot of literature on it, but is it really what you're going for? I kind of imagine you as kinda more interested in the kind of scrappy independent group of developers who have to, like, make something work amongst each other and who are serving kinda more lower level everyday users.
Yeah. So I think I've framed it in terms of, like, the and what I'm if what I'm hearing you, if I'm hearing you right, you're talk you're saying that looking at the level of these businesses against each other might not be, like, the same focal level that a a Nostrum approach might take. And I guess I would say that I framed it like that because, it's kind of like a simplification, but I would say that there is more kind of lower level stuff happening, especially in terms of of, like, the dynamics between the researchers who actually use and sometimes contribute to these frameworks. Like, I glossed over it. Yeah.
I'm super sorry. We're just out of time, and I don't wanna, lessen next time, but please do continue conversation offline in the chat, on the Slack. Sorry. This is the downside of lightning talks, but, Nick, over to you.
Thank you, Max, for that awesome presentation. Can everyone see what I'm sharing on my screen?
Yep.
Okay. Cool. First time presenting this, so we'll see how it goes. And and thanks for keeping, time aggressively here. You can reach me, Telegram or Slack. I'm here to talk a little bit about composable reputation systems. The idea here is oh. Oh, shoot. So the goal is I see a world where every agent so individual people, DAOs, and out, AIs can measure their whole world, the other agents in their world, through personally meaningful reputation systems. And reputation systems have been a big challenge over the last five years for everyone who's been working on DAOs. It's like something that's always been a promise that's been interesting, but it's never actually come to fruition. And I think a, it's hard and b, people have been trying to do it objectively rather than subjectively. And the reasons why it's hard is actually designing a reputation system, to begin with. It's just challenging. Like, you You know, what categories do we include and what metrics do we associate with those categories? Actually writing the you know, turning that design into code is hard. And then, it's hard to get a good one because if we design it top down, we end up with something complicated versus, allowing something that that, complex and antifragile to emerge, by iterative development. So the solution that I'm proposing is a no code reputation system builder, which makes it easier for people to make their own reputation systems and then a place to discover all the existing systems where they're being used that you can fork. So that's where the compostability aspect comes in. So high level of how this might work. I think that the primitive of a reputation system calling a block is the connection between one category and a metric or multiple metrics. So I'll get to an example in a second. You could have multiple metrics and each of them have weightings. So for example, if we wanted to have a reputation score for people who've funded public goods, we might pull the first metric as total funding contributed to Gitcoin projects. That's obviously missing a lot. So we wanna be able to iteratively develop that over time. So say we we're gonna someone's gonna fork this and make a new one, or or we're gonna incrementally develop it as a dApp. So we might say, okay. Not just the total funding contributed to Gitcoin projects, but also the number of projects funded and also the age of the first funding contribution. Say, okay. Well, it's not just Gitcoin. You can imagine how this develops over time. But the the idea here is that you have a no code system that allows you to find this, use it, fork it, and, modify it. By the way, a category is not just, metrics and weights. It can also be other reputational blocks and weights. So you start to see this you can imagine a system where people are developing important reputational blocks, combining and remixing them, adding new ones, and, kind of scaffolding an entire, complex infrastructure of, reputation systems that can be used by individuals and by DAOs. So the the key component of this is that you get a you get to true composability where you can fork existing projects and reuse them. You can modify them. Real inspiration here from, say, like, the token engineering commons bonding curves that Griff presented, last month or two months ago, and and the way that they adjust parameters to get to something that is, more stable and more aligned with the goals. And, obviously, like, the, you know, remixing from creator economy idea. Just like, okay. I wanna take that, add it to this. Or my personal value system, it calls for a reputation algorithm that incorporates public good funding, but also, people who contribute to academic research and also people who are working right now in DAOs and and are on the front lines of doing experimentation. So, like, how do I how do I use the, building blocks that are out there or start to create my own that I can get to, a meaningful personal reputation system that I can use to evaluate all the people who I interact with? So a little bit of what this might look like in practice. This is a tweet that lives rent free in my head for the last two months. The key is at Gnosis Guild, and she tweeted this, and it was like, oh, that's so interesting. DAOs as header heterarchic networks, ranking them in multiple ways and thinking of an adore an org as adaptable and having cultural practices and distribution of resources as the main focus as opposed to a structure that's the main focus. And what a reputation system does is it allows you to propose all multiple different and maybe even on a use case by use case basis or a workstream by workstream basis, structures that are ranking, you know, who who's the best for something to let them into the DAO, to let them into a subDAO, to try to find the best designer, the best writer, or the best editor for a particular project. I think there once we have this, we it could even be used for governance. Right? Instead of coin voting, it we could be, doing liquid democracy on a reputation weighted basis, for particular reputation systems. Ideally, every single agent, so each one of us, every DAO that we're in, can publish one or more reputation systems to evaluate other agents and see if they wanna collaborate, work together, invest, etcetera. And then in DAOs, I imagine a world where we the a DAO can use its collective decision making power to decide on which reputation systems the DAO as an agent wants to publish. So a world where individual members design new systems or and and forks of systems or forks of existing DAO systems, publish them as proposals, the DAO votes, and then the we're getting to that complex version of an antifragile reputation system as opposed to a human mind designed it, singular reputation system. And then I have this really cool, wireframe. This is kind of I wanna go to questions, so I'm just gonna show you this and not the other slides. This this is like a profile of my personal reputation that I'm publishing. So not only am I not only do I evaluate others given a reputation system that I have, I can share with the world the reputation systems that I care about and the way in which they're evaluating me. And if we have any, gamers in the house, you might recognize some of these UI components. But the idea is, like, a truly decentralized tool where people can, share their reputation algorithms, choose which ones they wanna be judged by, publish that information to the world, and hopefully, democratize access to social credit scoring in a way where we we don't end up with a dystopian centralized institutional evaluation of our data. Thank you so much. I'd love to hear your questions.
Thanks, Nick. Could you actually go a little bit more into, like, how exactly so what makes this composable? Like, what are the sort of module elements that go into sort of build?
So so imagine a smart contract that has a you have one category like public goods funding, and you you're saying, hey. Here's a reputation score that measures who's the best at public goods funding. It the only metric that it pulls from is the total number total amount of ETH sent to get coin projects. That that is not that good of a reputation score, but it's out there. Now imagine a smart contract that iterates on that and and forks it and changes the changes the metrics that are associated with the output reputation score. That the composable in that it can be forked and modified. And then it's also composable because I personally care about public goods funding, but it's not 100% of reputation that people might have with me. So I might say public goods funding is 25% of my of reputation that you could get with me. But the other aspects are other building blocks, and combining those is is the composable nature of what I'm referring to.
Okay. I see. So it's like it's like, you're aggregating it's like expert voting kind of thing. Like, there are, like, different contracts, different kind of indicators, and you can kind of, like, aggregate them to produce your own kind of, reputation metrics that kinda work for you. Right? So you Right. Somebody you can detect this person's reputation from your perspective.
Exactly.
Okay. That makes sense. Cool.
Any other questions or comments before we wrap up? Feel free to message me if you wanna talk offline. I'm on the Slack, and you can also message me, Twitter, and Telegram. Thank you.
Thank you so much, Nick, and thanks, everyone. I know we're exactly at time, which is itself an achievement. And, yeah, thanks for making lightning talks possible and everyone for presenting. I really love these, to see what everyone's up to, and the next ones will be in March. I put the sign up link in the chat. So looking forward to that. Thanks, everyone.
Thanks, everyone.
Bye.
Nick you just left.
Oh, no.
It's okay.