From Party Tricks to Production: How Web3 DAOs Advance AI | Jan Gornzy - Zircuit

[Music] of course you all right well everyone please give another round of applause for my next guest Jordan Craig thank you thank you hey folks thanks for rolling through um basically I'll do a quick introduction to myself um I'm Jordan gray and I am co-founder and CMO of public AI Where We Are are making it Equitable for people to add their experience to AI uh to make AI models more exceptional um today what I'm going to do is go into a lot of the more detail of how exactly we do that and how my background in building Dows got us to be able to use web3 Primitives in new and interesting ways to meet those goals for our paying clients the QR code that's up here will allow you to go ahead and try our system that's live today so you can uh join up and play around with it and check out what this is all about uh it's all real stuff that I'm talking about so what problem are we facing when we're talking about using Ai and using web 3 to enhance AI um when our clients come to us what they've done is some kind of large language model experiment and what large language models are great at is spewing out a bunch of Tax store images that are semi- believable it's this latent space that is is getting Tighter and Tighter but it's all very generalized and when our clients are trying to apply AI to their businesses what they're missing is something that is focused and able to deliver uh reliability and also cost Effectiveness when applying AI to specific business problems why is it that they can't get it to that level of reliability well it's because AIS llms are very generic the reason they can do these party tricks and generate text and images and videos that look so believable is that they're so generalized so it's very easy to do something that looks plausibly plausibly reasonable um for most use cases but then when you ask it to do something focused and specific it struggles with being able to nail that reliability why it's the data the data behind it is trained on the entire internet is trained on synthetic data these generalized use cases are a perfect fit for that but when it comes to actual business applications folks need specific focused data um and to get that data centralization is not the best way if you go to somebody like Reddit or Google and try and get specific data what you'll get is a giant data set that's not what most businesses need to get their AIS reliable in production so what web 3 can do is go to a large set of workers and get specific data relevant into those business needs and that's exactly what we're doing at public AI we have a community of over 700,000 data Builders hence the uh construction vest Vibes we're all constructing building the data together um and with that we get awesome diversity we've literally got people on every continent um from all different socionomic demographics uh the diversity is amazing the quality is also very diverse and that's less amazing so how do we get great quality out of these data workers that's where the web3 Primitives come into play so we use a combination of Stak slashing mechanism consensus voting and verified reputation to ensure that we get quality out of those workers through a pipeline to go through those one by one the way that the staking mechanism works is there's two stages first we go and Source all of the data um it gets uploaded and that can be any modality images text videos uh audio anything and after it's been sourced we then do a stage of verification where voting happens and people are judging whether or not the the data that got uploaded is valid so to participate in the uploading and the voting the users have to stake to say I'm committing to making a quality contribution here I'm going to be adding more signal than noise and when they do that they get their stake back they get paid and their reputation increases so that the next time they go and participate they actually have to stake less because they're a trusted actor in the system of course if they don't add quality then the exact opposite happens they get slashed and yes that's kind of harsh but that's the exact incentive mechanism that needs to be in place to ensure that quality comes out on the other side of this whole process that reputation that gets built up also puts our users into different different voting spheres or Stratus so when you first join the system you're considered a scout you're out there exploring getting your feet wet um and the way that it works is when you're on the Scout Stratosphere it takes a lot of Scouts voting together to be able to say data is valid or invalid and then after you build reputation you become a guard and at the guard tier it takes less people voting together to be able to pass data or say that data is bad whatever the consensus is and then at the very top there's the oh my HDMI is not working yeah um at the next tier there's the judges and those are basically the clients um other very high reputation individuals who have been longtime contributors in the ecosystem um and ourselves as workers uh in the company being able to kind of administer at the very top um with just a few votes to say if something is good or bad now you might think it sounds very inefficient to have a large group of people all voting and not being able to pass or um pass or deny a piece of data but by having all of those people voting along the way and then finally up the the pyramid having people being able to say yes this date is good or bad we get strong signals about if these contributors are making uh real good um contributions to the system or not so this is essential for building the reputation of the contributors in the system in addition to that we also put out surveys to our users to be able to identify which ones of them are experts in different areas uh for example if somebody is a pet lover um loves anime is into handbags knows math or science uh speaks a specific language uh we do uh surveys to be able to see which one of our contributors are the right ones to put uh campaigns out to to begin with and the big question that we had to answer for ourselves and also most people ask us when we explain the system is do people actually show up putting a Stak slashing mechanism in place is a very high barrier to entry it's hard to get people to vote in the first place nobody likes governance and it turns out that the answer is yes um when we put out campaigns uh we have you know some fraction of those 700,000 workers showing up um and then because of the stake slashing mechanism we get a smaller fraction participating and that's exactly what we want is the people who are dedicated to providing quality um if somebody's just doing random Behavior clicking around trying to farm without providing actual value then they wind up paying us for putting noise into the system but the people who do show up and provide quality they get more rewards because all of the people who got slashed um aren't getting rewards so the way that the economics work of the system um means that the real workers get paid more and the farmers get paid less we were able to verify all of that through building this first um in an MVP way offchain using points so we used a point system and have that deployed today when you go and join a campaign you get paid in usdc or other stable coins and points and then from there um you will be doing a tte next quarter so for everybody in the audience there there's your when Alpha uh next quarter um and after we release our token then we'll be doing this whole system uh purely with our onchain token but the important part of building this first with points was that we were able to identify which components of the system needed to be edited uh very frequently and in real time in response to user behavior and those are the values that we're baking into the smart contract to be able to tweak in real time as we're seeing people uh interact with campaigns so in the end the big question and vision is is this a killer use case for web 3 and we think so as I mentioned earlier I've been building in web 3 for over 5 years uh I built Dow infrastructure at near protocol where I worked alongside Ilia pukin and we had over 2 billion in assets locked up in uh dowo infrastructure there and through building that uh Dow infrastructure also helped hundreds of projects launched have seen so many different use cases for this technology and in a lot of the cases the people who were reaching for Dow software they were grabbing tools that were an approximate fit but weren't exactly great for what they were doing what's really exciting about what we're building at public AI is that we have paying clients in a very specific use case and we're taking those same web 3 Primitives and applying them to a very specific problem which means we're able to tune the software to those exact needs so one of the things that um frustrates me about the web3 space is seeing people kind of do the same thing over and over again uh deploying another fungible token uh doing another crater coin economy when we know the exact shape it's going to take we're taking Primitives like staking and instead of just saying hey we're going to offer you yield we're using it to verify the quality of contributions in our system and this same thing with the voting mechanisms and the reputation we're not doing kyc to identify if you're allowed to go and get an airdrop we're verifying your skills in keeping track of that onchain and that's all Anonymous so you can be anywhere in the world we don't care what we care about is your expertise and your history um and what value you can bring to the system and in the end when we're talking about how to get like value derived using web 3 and what that means for AI um the only way to get truly aligned models is through aligned incentives and when we have top down uh values driving uh where models are going that means that we're going to have privatized interests in control of those models and there's not going to be any place for humans there so we see it as imperative and essential that we find routes for humans to come and bring their expertise their interests their values to these systems so that even before we get to AGI or ASI we're able to see ourselves reflected in AI systems and our values reflected in AI systems um so that we don't wind up in another situation like we had in the web 2 world where we're basically being farmed and all of the values being extracted from us so in the end through these web three mechanisms we're able to get very high quality data from a very chaotic uh self-emerging self- selecting group of workers over 700,000 people are showing up contributing data verifying each other's data and enriching AI with their Human Experience um some of the clients that we're working with these days include um somebody who's building World coin for dogs so we have people going and taking pictures of dog noses um we have a real-time translator client they've they're doing something like um like Babble fish from Hitchhiker's Guide to the Galaxy so for them we're recording a lot of uh audio samples and local conversations and very Niche dialects and for clients like Eliza OS and other AI agent clients were helping them make it so that when all of the messages are coming into their systems they don't have to spend as much on saying whether or not a message is worth replying to or not which uh turns out to be one of the most costly parts of running those AI uh agent models and in the long term where are we going from here what does this mean um one of the big Moes that we see in the AI space is data and that's why we're solving this problem compute keeps on getting cheaper and more available uh companies like deep seek are making it even easier for everybody to reach for AI and for us that is awesome because it means more and more businesses are going to be thinking about how to apply AI to their businesses what that means is that well yes there might be some jobs being lost to the efficiencies of AI there's also more room for solopreneurship and smaller companies to be able to do more so we're expecting there to be a greater diversity of initiatives companies um and smaller and medium Enterprises that are going to be flourishing with access to these tools and we're prepared to be able to help them figure out how to apply AI to whatever their specific business needs are one thing that you can kind of think about uh that drives this point home is the Indie gaming scene where through the diversity of small Studios being able to develop products that appeal to Niche audiences you get a great diversity of products that allow you to you know find a game in whatever weird Niche you want and that's going to be a wrap um the call to action is to go check out the platform uh data Hub is live today uh you can play around with it it's a product that we're iterating on and we'd love your feedback so would love love to see you in the community it's a lot of fun um I'm there in the trenches uh looking at dog noses with everybody else and also if you are a small company a startup and you're thinking about how to apply AI to your business would love to talk maybe we can run a campaign test things out uh see if we can get you what you need and then besides our uh tge and Q2 that I mentioned we're also going to be doing an nft collection very soon and that's going to be a wh list for our deepin uh devices which will allow us not only to collect the data from people's phones and desktops like the audio uh video text Etc but we'll also be able to collect things like brain waves and other uh dite type data with our um dpin devices going forward and that's it so I ran through the slides pretty quick that leaves us with time for questions if anybody has any uh small audience but uh we got a mic right here if anybody has any comments or questions I explained it super Well everybody's happy nailed it great well thank you so much thank you Jordan Craig