Fireside Chat I Illia Polosukhin & Kartik Talwar I Operating System, Open Web | Pragma Waterloo 2023

foreign [Music] last Talk of the day and then we get to talk to all of our speakers outside with a bunch of drinks and I know all of you are been very patient here so without further Ado I'd like to bring on our next speaker Elia Elia is the co-founder of near protocol and some of you have seen near protocols some of you've heard of it some of you have heard of what Eli has done in the past before starting the year we'll cover all of that and what the feature looks like a multi-chain world in the next 60 seconds welcome oh well we'll cover starting that there we go don't worry I'm also tired but please welcome Ilia from near it's been a long day I know I was talking Mr Carter yesterday I'm like you're gonna be on stage seven hours like are you sure this is good for you he seemed like a good idiot last night well I want to cover a lot of the blockchain stuff I want to also cover a lot of how we should think about near versus ethereum versus all the other teams that are out there and what that world looks like on scalability but uh I want to start off by asking you what did you do before starting the air and uh and you should tell the audience kind of how you got into the space and what made you start near yeah so my background is actually in machine learning and uh uh I got kind of excited about it very early on uh I think somebody asked me what's my like favorite movie from uh kind of teenage years and I responded artificial intelligence and so uh not a lot of good IMDB ratings on that one but it was a good movie it was a good movie and uh definitely kind of led to a lot of thinking from there and so um I've been working for about 10 years in machine learning I've done a bunch of like projects uh kind of across you know everything from cancer research to Insurance to finding oil and then I joined Google research to work on natural language because I actually think natural language is the best way to represent knowledge um and apparently that's true now that we I think I had a chatbot here would agree yeah and so at Google researcher worked on if you search for stuff at google.com and saw kind of short answers for this like our our kind of team was powering a bunch of that everything from like lists to actual like short answers um and I worked on tensorflow which is machine learning framework gather through the kind of whenever I saw open source I'm like I need to work on that um and I ended up kind of through being uh in kind of it was actually a funny story because the whole attention is all you need came from lunch so it was a launch conversation okay that's a that's a story I don't think I've heard before either so yeah some of all known story but yeah it was a lunch conversation so I mean the context was uh we were we're building a lot of kind of recurrent neural networks so uh if you're not familiar with neural networks it's kind of you're processing one thing at a time and so it ends up being very slow and very unstable so it's like easy to go really hard to train and so in result we couldn't launch anything like that in google.com or translate uh because it was too slow to kind of do inference um and so instead everybody was doing just bag of words so we like which is just you know you literally don't have a sequence of words in your model like it's not considering sequence of words you literally like just take a sentence split it into words and feed that into a model without any order uh which was working surprisingly well I didn't notice on Google I don't think you would notice anywhere else and so and so the idea of Transformers originally was uh so we used it like there was a concept of attention which was um kind of pioneered by like a year or year and a half before that and I was like well with it if you use a bag of words and attention together like can attention actually recover the sequence and and and instead be the power instead of the recurrent neural networks so that was like a launch conversation and so I went back kind of you know hacked something together and like it was not not noise and so then like uh me and Ashish and like others like we start kind of trying to bring it to life and see if it works that's awesome so it turns out that attention was all you needed exactly uh for those in the audience um he's being a bit humble attention is all you need is the actual invention of Transformers which is what powers all these llms and open Ai and chat CPT and everything else he's one of the authors and inventor of that concept and uh and that was almost five years ago yeah it was 2017 uh this was like uh beginning six years ago yeah and so and seeing that come to life now and everybody kind of finally realizing what you can do with LMS is insane so inventing Transformers and then something in the middle and we have near how'd that happen yeah so uh so paired up with my co-founder Alex kiranov and uh kind of I wanted to do something like we had kind of Transformers I always wanted to do code synthesis uh so far so it's not familiar it's what GitHub gold pilot is and so we kind of started near AI pretty much to build GitHub co-pilot five years ago five years ago without Microsoft money or GitHub data uh and so uh what we did was uh our competitive programming background and so we got a lot of data from that and then we had all of the students around the world actually working for us adding more data and so we uh and the challenge we had was actually paying them because we had students in China in Russia and Ukraine and kind of all those countries which have monetary Control Pro like the US capital control yeah Capital control you're sending money from us to this country so like complicated some students didn't have bank accounts uh and so the crypto was like hey we have this global payment Network that everybody's talking about and this was like um you know beginning of 2018 um like can we use that uh and the answer was no you cannot because it's like way too expensive for what we're doing way too complicated even for developers to figure out how to use it it's like hey before you can receive money install metamask and fund it you're like wait with money we paying people to to do stuff can we like refund it for them it's like no you cannot like they need to have their wallet it's like so so that was all kind of uh drivers like this doesn't seem like the way it should work the idea is great A lot of concepts are great but like it doesn't work and so we started digging in and trying to figure out like how should this work who's working on this what are the kind of ideas in the space and so that's uh when we met vitalik we met Justin Drake we started also meeting all the other kind of protocols as well diving in uh how it works and at the end we haven't found anyone who's trying to make set of trade-offs that we believed uh right which by the way very similar to what uh kind of flow is uh uh thinking about which are how do we make it really simple to use for like people who don't know anything about blockchain and shouldn't how do I make it really simple to build and kind of as available to anyone uh to any developer without needing to learn like new Concepts and then how do we make it really secure and and scalable to billions of users right so that like it actually linearly scales as more usage comes in um the the kind of analogy I like to use is that you know Netflix never like asks you to pay more because like other people are watching the same movie so uh so like it should just scale underneath and Uber does well maybe they need the better that better like I think they were working autonomous driving that was supposed to solve that no so anyway so that was kind of the the origin of near uh we ended up doing a really if you haven't seen a whiteboard series which was a really cool um way to learn about what everybody else in the space are doing and so we interviewed everyone from uh you know Cosmos from ethereum from Solana polka dot so there's like over 50 episodes of filecoin is there as well and really diving into like how technically the their protocol is working and really kind of what are the concepts and what are the trade-offs um and so that gave us a lot of understanding and as we were designing near we were kind of understanding very well what is the trade-offs we're taking and also uh kind of how that compares to the whole Space you know I want to kind of interject here for a second because this seems like a such a casual thing that you're saying but three years ago we had to invent a lot of these things and hope that they worked in production that we they work in theory and uh that's going to apply to anybody who launched a blockchain anybody who kind of played with different consensus algorithms which is different ways to think about scaling so it feels like oh of course it would work or this is a given now to everybody who's coming in but this was crazy five years ago people were still skeptical of E2 and sharding and and we've kind of seen the iterations on all that too so maybe before we kind of dive in we'll work kind of some of the things that you either ended up inventing or iterating on from existing computer science problems that kind of led to near being what it is now yeah so I mean to kind of the underlying question pretty much is one computer like the blockchain usually realizes that there is one computer uh that you know produces a block and then everybody else validates it meaning like they kind of replay the blog execute transactions and get to the same result as everybody else and so in reality you ended up with capacity of one computer so you cannot do more work than what one computer can do and this is without consensus overhead right this is literally like if we assume you know there's no networking lags there's no like need to wait for agreement it's just like you execute as fast as one computer can do and obviously that's limiting and uh there's other other sets of uh kind of limitations then you add consensus which is well if we need to wait for an agreement before we can produce next block that's extra latency and in turn less uh uh less throughput to the network and then on top of it you have the problem of uh kind of VM execution and actually the real problem is storage because if you want to have authenticated uh kind of storage right and what this means is in blockchains you can verify that this piece of state right your balance you can verify that indeed you have this much user or whatever token in your account and there's a proof for that and it's kind of consensus proven this and so to do that we actually store it in so-called kind of authenticated database right which means there's a proof for each piece of data and that is very expensive meaning if you have you know a million records you actually need like 10 writes and 10 reads for any single piece of data to write or read complexity for computer science students here and so so that's a very expensive right because there's you know if you have a transaction that touches like 100 pieces of state that's you know can be like a thousand reads of disk which is really expensive even investors ssds and so so all these are all the kind of problems if you decompose a blockchain this is you know this the storage VM execution can uh uh kind of consensus and then like the fact that you only running on one machine and so the way to solve this is you know kind of work on each of these layers and figure out how to remove the bottlenecks and so starting from the top the um kind of the single machine thing is how do we have multiple machines running in parallel executing transaction parallel ordering transactions like good intersection in parallel then how do you pipeline consensus in such a way that it doesn't have a lag right so ideally at any point of time somebody is producing next block and exit and while while somebody versus the next block you executed in the previous block right so that you don't wait like while you execute to produce next block and on kind of VM level how do we optimize performance it's as close to Native and on Storage level how do we kind of remove the bottlenecks and and when you know when we talk about rights how do we pipeline right so we don't need to wait for them and for reads how do we have ideally in memory or at least like flat data storage that we can read from so this is all kind of pieces right and like from there you need to like really build a thing that works combining them is not yeah yeah complaining yeah and like the specific the sharding part like the running things in parallel is hard because usually when people come to blockchain they expect this Atomic execution right they expect that if I send a transaction and it touches a bunch of contracts or apps if something goes wrong the whole thing rolls back and you cannot achieve that if you have multiple kind of parallel execution happening right it's the same kind of going from you know single threaded application to a multi-said application now you need to think about kind of logs and complexity around that and so that was one of the core trade-offs right which uh you know with layer twos with uh polka dot like the decision was different the decision was like well why don't we have still this like single threaded applications uh but just but we're just like have a bunch of them and we and then we figure out how and the users and developers figure out how this should work right and like from our perspective you know I started with simple and secure and secure and scalable right the simple was the important part and this is not simple this is like makes complexity surface to the user and saying like hey do you know which layer 2 you're on or app chain you're on or power chain you're on not zone or whatever like you need to know like which security it has which like you know who's running it is it is does it have all the ethereum and account to actually submit blocks like you should probably monitor that as a user and so so that's the complexity that like uh kind of the other trade-off has and we we wanted to remove that and user not think about it and in turn it meant that we have each account each contract to be running in parallel as well so we don't have this atomicity on a transaction level we have it on an account level I mean that's a very hard problem to solve with locks and just making sure this is globally consistent we won't get into how you solved it because I and so that's there's a video on YouTube yeah we'll send you the link uh but obviously this is now we know that it's it's working and it works great so let's go into understanding what is now enabled by these Primitives it's becoming more accessible from whether it's from the nearest perspective or just the wet through ecosystem overall what are kind of some of the themes that you're seeing that are now evolving and becoming more and more possible yeah I think kind of probably the to finish the story right so we started with layer one we really kind of focused on making it simple to use so we did a lot about the kind of extractions kind of early on as well making our accounts extremely flexible and Powerful um it's like everything from you know multiple keys to turning them into contracts back into accounts somebody just launched turning counter to nft and back as well so you can do all kinds of crazy stuff with them uh we have webassembly so you can run rust JavaScript python whatever you want as a smart contract um new compiler can run and compile for that as well Cadence and so and we have sharding right to actually scale and so the kind of the next step of that was and like it was funny to I mean kind of look at Power Loom before because yeah it's like the next problem is okay well now you have all this blockchain but that's the data is in completely different shape right the usual database problem data is always in the wrong shape for your application and so uh how do like we kind of worked on indexers and different Frameworks and apis and sdks on top of it to really power it okay and now that we have that the next step came in which is well there's a lot of smart developers who build interesting applications there is people who are actually creating really interesting uh things but they don't have like a full story of like product onboarding kind of all of the pieces together right they usually don't think about it uh at least yet and so how do we power that more and kind of the evolution that we came up with we call it blockchain operating system it's really an idea that right now we have kind of blockchains which are really kind of almost like Hardware right they have like a very low level uh apis they kind of provide a very low level uh you know Primitives of asset ownership of uh kind of um business logic that you can program for and as well we have a lot of them so we have like fragmentation as well across but as a user you don't really care about that you just want to go like you know use the map trade some nfts you know play some games uh you know swap some tokens not caring about slippage and Mev and all this stuff right and so so that's a disconnect and as well as like as a user you want a product you don't want a you know one app that gives you like one feature like uni swap like you want a product that gives you notifications you can search for stuff you can like interact with it you can find dashboards you can do all those things in kind of one place and so blockchain system is really connecting all those pieces together it abstracts kind of blockchain complexity with all the data layer as well as meta transactions as well as uh kind of indexing and things like that and then on the other side it creates a user Journey right it unboards users we call a thing called Fast off five seconds you sign up as an email again named account you can user apps don't need to pay fees you don't need to figure out what like 0x something means you don't need seed phrases like all that is abstracted out uh you have discovery which is decentralized front ends so this is turning what smart contracts did to Services doing that to front ends as well making them decentralized really composable kind of living on blockchain but in turn being rendered and kind of creating this very powerful developed environment in which you can kind of navigate from one app to another inside the same uh kind of platform and so for those who like you mentioned Facebook and Zynga before like this is what Facebook was kind of before it's like a platform in which you can build new apps really easily and then they kind of close it off so this is an open version of that like everything is open source you can Fork any any other application as well and build on top of it and uh you can spin up your own version of this and kind of showcase how it works you have search notifications and kind of all of those Services as well that like users are used to and kind of packaged all in one kind of really like powerful platform that everybody can use and so that's kind of where we see this evolving is really like we still you know truly preserving the principles of Lab 3 of user ownership of decentralization of anyone kind of being able to run their own version of this as well but the applications the front ends the data everything is common and so it doesn't matter which Gateway you come from it goes through near.org or boss.gg or potentially in some Brands gonna support this you you can actually navigate to the same apps like it doesn't matter where you start and so we're kind of turning around this problem that decentralization meant you have no distribution because nobody can find anything to really decentralization actually means you have more distribution because now every channel that installs this operating system can navigate to say Maps yeah and so that's why again like going back to the subredding system uh analogy you know as you get more installations right you now have bigger platform for this applications to be used and the important part is like it works as any blockchain right you can you like we have people uh running on layer twos on this idiom uh folks are adding other like Cosmos chains as well and so it kind of spans like the specific you know also I call it a religion like instead of like hey you need to convert to my release whichever culture you're part of yeah at the time you're like hey this is just solving people's problems right for defy this is decentralized front end such that you don't need to run and maintain front end and be on a hook right for uh other applications like hey this is a quick Discovery because all the users right away can access this and not need to connect right you have you know social features that already embedded there you have kind of all those pieces really coming together in one platform so and like to your question like this is where we see this evolving is really like this kind of becoming more of a ubiquitous platform that leverages different website Technologies under the hood but as a user you're just navigating it and using it that sounds like a pretty good pretty good pitch it's almost like I've practiced like 50 times in past three weeks but yeah you may not realize this thing but we did plan a lot of this so not the talk uh we're not that coordinated but how we get to this conclusion why this talk is the last one uh part of kind of the goal for today for pragma was to kind of tell people about what you should be actually looking forward to in the next couple years and if there's one thing you kind of take away from this it's the fact that things are going to be on chain and that's just you need to accept that because that is great for everybody you get so many brand new features and you get to optimize for things that we are currently suffering from a lot of that is distribution and portability of information which is what decentralization actually solves because it gives you the ability to exit we kind of start off in the morning with talking about what's actually happening on you to get a lot of users to get into crypto for the first time that's with rabbit hole we saw a lot of information run okay now that you have some information about a user how do we actually tie all of that into actions uh that was tally especially with power loom which is the last demo it was okay now you have people's information you know who the users are you want to be able to tie all of that into actual data that they're producing to be able to reference that later and if you have all these data pieces you have the networking pieces you have the user information you kind of get all of that together into packaging that into an actual front end which is what people usually interface with and having that be reliable and consistent and not be beholden to one single web services provider taking that away is it a beautiful thing because not only is you kind of talked about the operating system I feel like the better analogy is that it's actually another open Internet that exists uh for anybody to also discover the same things without relying on one single DNS resolver to decide one day that they don't care about this thing anymore and effectively send censorship can be mitigated by having multiple copies of full-on stacks whether it's contract data applications and your JavaScript bundles all in the same place so to me this is personally super exciting and kind of my question to you is well seems like from what I understood all of this is right now on near and you kind of said it's actually not specific to near so could you tell us more about how does this actually apply to any other blockchain can I run a unit swap front end there in that world like where's that interfacing with how does it know which one is ethereum what's near tell us more about how this works with the rest of the world yeah so I mean the general principle right there we're trying to follow is that everything is is open source and everything is kind of componentized and so you can pick whatever set of components you want for your own use case and so in case of you know you wanting to um what would be some components here so the discovery is the this like decentralized front-ends component right so this is where you can build this front ends you can build them for uni swap which already exists you can build them for you know uh Galaxy for uh Ave I just saw it actually uh somebody built it somebody built for one inch I just retweeted that uh I'm wearing one in soccer yeah and so so you can build that for any network that kind of supported by the web engine the the thing that runs it and so the idea there is yeah like it works with any unblock chain already the for example fast toss right and uh is right now works on near and creates near accounts but the next stage will be to kind of also create so we call them like remote accounts and other chains so you can start like as you onboard through this funnel you can start interacting on other chains as well directly from the web app directly from the web app the same experience you don't even know that your interactions are the chain right same with kind of indexing framework you know kind of going to be integrating other data indexers from other chains so that such that as developer you don't really care again on which chain it should be you don't need to like change apis or just say like hey get me data this kind of data from this contract on this chain and it just pulls it up right and like so underneath kind of all those components need to just like play together and and like be expandable and support any chain nice how how do we think about data Integrity in this world like how do I know that the units while front-end I'm using is not malicious or the other person doesn't update it after how should we think about that so the front itself I mean it is stored on chain so that's stored on near protocol and it uh is signed cryptographically signed and so when you retrieve it you have actually approved from the blockchain that it's indeed this front that you're looking at so any future updates to that code will be stored on chain yeah so you can if you pull it up right now near that organ like you can go to develop mode and see the source code of the page you're looking at and you can go to history and see all the changes as well that's awesome and you can like if you fork in it or or using this in your own for example you're like hey this unit swap looks cool I want a mega swap right which has unit swap as part of it you can link to either specific version or the latest version as well or you can Fork it and link to your own version it's awesome I well okay so you kind of just there will be a workshop tomorrow as I show you without like me trying to explain no more live demos um well this is super cool because you've effectively made it easy for things to actually work on all devices too because now if I have portability on my desktop and mobile without having to have the same key copied I get access to the same Maps is that is that a good way to understand it exactly yeah I mean potentially this can be like rendered in command line like for those who are like emacs fans like you can you can go into that mode as well like the idea is that the kind of the essence of the application is now like stored and and the way you render it can be different right and so uh the way it can be rendered into a web it can be rendered to mobile it can be rendered command line it can be rendered into LED lights right and uh on in turn as well we actually just partnered with zorbit for those who are deep into the rabbit hole and so orbit is like kind of self-hosted like when you actually have a server yourself um the OG religion in this yeah all G religion yeah and so so you can run that thing off orbit as well and kind of have all your front ends making sure they are all your own and nobody can affect them right so I mean that's incredible actually also if you haven't noticed I am learning live what what the features are which is fascinating for me um I want to ask one more question on this thing when I move on to a couple other topics that uh that I feel like are worth kind of talking about because it's a shared theme on all the other interviews um how does stories work in here like that sounds super expensive to keep all this on chain what's kind of the the magic trick uh well the this the chord trick is using history of the blockchain which is way cheaper I mean it's the same trade parcel divs of the next uh data well we actually we store right now the full but the history is stored in in the history of the blockchain so same as like so that's called Data more data instead of um but that means there's even more data than you would want well so a like because nearest sharded the chassis generally the storage is way cheaper than understood like and in general the storage is not that expensive as was mentioned uh and so we don't have that limitation that uh a single machine uh yeah it's not a global everybody has to sync the same storage and I reckon is like for history like we actually store it in the history of blockchain as well which is pretty pretty robust and reliable and at the same time way cheaper so like you know if you think of GitHub right GitHub actually has all the all the history of all the rebos on top and they're replicated as well right and so like the cost there is I mean not maybe like five seven eggs difference was what uh near does so and it's not that it goes down over time it goes down with the cost of yeah Hardware that's super cool um I want to switch gears a little bit on just scalability altogether my question to you as somebody who is not in the ethereum OR directly the flow ecosystem from at least today what are you getting your perspectives on layer two's um and if I were to kind of ask bluntly are layer 2 scaling from your perspective so I I mentioned this right uh kind of the approach to scalability is all the same across everyone right it's like we need more machines to run stuff in parallel and that's the only way to do it like just physically well sorry the other one is bigger machines uh more computers either a stacked up or yeah yeah and so uh so what layer 2's right now they they're done this first step which is like okay let's pull out the execution away from like consensus right so they said like Hey we're gonna run this execution machine and then we're going to settle on ethereum which just adds latency for the ethereum and it doesn't really individual rollout doesn't scale yet right so if they need to scale the roll up they will need to actually Implement sharding on the roll-up side is everybody says well let's run another roll up and although it is indeed more stuff running in parallel now it just becomes more complicated because and from what we've seen so far it's not like people say hey we're gonna run you know are they here and unit swap there it's like we're gonna run all the same stuff on every single row up uh and so again not clear what we're scaling there so from my perspective right now like kind of there was a first step made which is like hey we like kind of moved the execution away from the consensus so we can easier do things but it did not really scale because and and in turn actually all these Roll-Ups that are now competing for Block space actually driving prices for themselves up and so it actually becomes more expensive because there are more Roll-Ups to run to tools like execution Wi-Fi for being included I see so so the solution is move everything to layer one well so there's two solutions either say okay we're gonna Shard roll up and so we'll they will need to build the staff way building on the roll-up side and then and then we'll need the ethereum needs to implement dunk sharding which is data availability which is again what near done uh and so so this will be deconstructed near into multiple chains and you'll have like multiple tokens and you know uh or you can like yeah collapse it into one integrated system and optimize it for latency and have like faster blocks and faster execution so that that's a trade-off and like I mean the serum takes a trade-off that that it's better to have people experimenting and kind of having more of this alternative service execution which is is a initial position yeah uh and we take the position that we can optimize more and deliver better latency and better throughput through pretty much making some specific design choices around like which VM is running and how those things are tied together but in turn give a better usability to the user and developer amazing I am going to go off a little bit tangent and then we'll come back to uh our last question for today um kind of from your perspective what's kind of the overlap do you see between blockchains and AI yeah so that's a good question I've never been asked this before that's a joke uh yeah so um I think there's like for on the eye side there's like kind of to understand the basics right AI is formed or the machine learning generally is formed from three concepts right there's data uh there's compute and there is a model kind of architecture and uh structure and ways of training so what blockchains are really good at is marketplaces they're really good at like creating liquidity for anything and so uh for each of these three things there is actually right now a very limited availability of each of them so if we're talking about data right now if you want to get more training data you need to go to scale AI talk to their sales person uh you know figure out how to get these things you know pay a lot of money and uh kind of start collecting data and you know eventually it's low quality and you you know you need to fire scale AI you cannot fire like individual person who's working on that and so uh the kind of the first step there is like creating a Marketplace for data for crowdsourcing this is actually what we started with um so then for compute similar thing if you want compute you actually need to go to Amazon uh potentially talk to their sales person you know get a get a commit uh for like seven million dollars a year if you want to do like AI properly and uh uh so that's again like something where we can have a compute Marketplace it is harder there's like a lot of caveats around how that is done and requirements there um but this is kind of conceptually and same for models right right now there's a lot of researchers you know investigating how to improve Transformers there's a lot of people kind of like if you go to Huggins hugging phase there's a ton of stuff there uh that people are experimenting with but there's no like marketplace right there's no way to like request hey I want to model fine tune for this or uh there's no way to kind of um also monetize models if you like do make it you either open source it or you close Source it and give an API so there is a opportunity there to create a combination of something that's private that doesn't like leak all your data to some company and at the same time uh give some uh money to the developers this uh model in in kind of like somewhere between open source and closed Source pretty much like a new way and this is where you know people in research in ckv Decay machine learning and other stuff but like kind of conceptually that is a kind of a powerful place as well no that's real good I think especially with like doing machine learning on private data without reading the data is is a in a way novel kind of thing that we were kind of just seeing at the beginnings of uh I guess one thing I was curious about any from from uh when I asked that question was understanding how do we actually think about authenticity of information because of course marketplaces work on the side where somebody's contributing to these data sets but how do I know if I get if I receive a message it's not generated from chat gbt or if I have all these defects or a voice message that just copies my voice after three seconds of input that it is actually me and not some someone uh faking something so I wanted to see if you kind of think about how does blockchain help with yeah so this is the flip side of of like this was like how to use blockchain to do Ai and this is like how do we use blockchain to make sure our society doesn't fall apart and and so yeah so kind of fundamentally our society runs on language right everything you do from you know kind of creating a form filing for your taxes uh you know lawsuits political messages everything you do is language and one thing that language models are really good at is at language and this is the first time actually in Computing where computers went from outputting something where you need a human to like interpret it and tell it to everybody else to the computer being able to communicate directly and this is where all the flaws of our society are surfaced it's not that AI is actually like breaking things it's this is problems that exist the broken system is highlighting it yeah it's just like the scale at which you can kind of now exploit it is tremendous right generating a million page tax return that IRS is required to read great spam the Judiciary System with lots of fake lawsuits no problem right create yeah fake voice messages from your friends uh like literally can be happening now so all this is like again exploiting the existing things and so the the point is we have cryptography and cryptography creates a very much like natural way to authenticate um and Link it to some identity and so then uh kind of let's say if we you know mandate that all the content generated needs to be cryptographically signed now we can actually know where it's coming from so right now voice messages by the way are not signed the whole you know all these just all of our communication yeah Communications some of your emails and so so the first thing to do is like we need to cryptography sign it which by the way if you could really assign it you need to associate that cryptographic key with something that's where pgp and gpg been like not great because it's been really hard to have this kind of registry of people and being able to exchange this keys well that's where blockchain comes in blockchain is really good at actually having uh kind of this reputation and so you just need a good account abstraction to manage that um and so this is where like where kind of all those pieces really coming together and again from our perspective what we're doing with blockchain running system is trying to power a lot of that is like social kind of we have a social component there which is fully cryptographically signed the front is a cryptographic signed and so having a default feature yeah and you have you can go to a user right now and see which everything from you know which transactions they did to which front ends they built to which posts they did Etc right so you can have now this powerful reputation being built like which allows they participate Etc to actually authenticate who are you uh communicating with because the other problem is sure it can be signed but then in case like obviously Bots can create accounts and sign things but you need to understand who is signed with and then the next thing that we need is actually Community governance we need I mean to the kind of point of previous speakers we need Community to be able to decide which things are you know kind of not appropriate in various ways right and and provide this kind of almost filters that users can apply on their communication on their you know way they read the internet Etc and so this is where like blockchain is powering all that as a kind of really core infrastructure there's a lot of pieces that need to come together like content Management systems and other pieces uh probably browsers need to support that like that's when you read an article you need to like validate that like hey this is not cryptographically signed or like hey the signature is from could be found on coindesk yeah or like this quote is not signed like do you you realize right now quotes in articles are not like confirmed small screenshots charges edited and so so all of that like we have now infrastructure to do that we just need to like now build it into the stack and and showcase to the also to users that they need to care about like I think the good example was the transition to SSL like the fact that was like you know people starting caring that their lock you know on their browsers not is like right there was a time to convince people that your website must be https and you can see the lock all of early 2000s and uh and now we need the same things yeah we need the same thing right now to happen which is like the browsers need to show like hey this page is not signed you may be reading like or like this book is not signed like books can be like fully generated you may be reading completely different thing under the same title right you wouldn't even know textbooks like imagine textbook that teaching kids like completely random stuff oh well that's a scary world and to be clear this is already happening that's been happening for hundreds yeah it's like it's already happening we're just pointing it out uh live so so maybe let's find a way to end this is live stream are people breaking Society right now not yet in two weeks it'll be up on YouTube we'll we'll Photoshop this later um deep fake it there we go that's let me solve the problem with the problem um well my last question to you is what I've asked every other speaker what does decentralization mean to you for me at the core is this um kind of low to no cost of switch cost right is like that the fact that you're not locked in into the platform be that like at any level right be that the blockchain self the block producers the uh kind of the wallet that you're using their PC that you're using like any any level of the stack it should not be like this is the only place you can get this and anywhere else you cannot and I think I mean that's why like front ends like is is has been actually a loss so like how do we like replace kind of every level stack is really important but kind of for me conceptually and like you know if we go into AI it's the same thing like it should not be that open AI is the only place you can you know interact with jtbd and it gets all your data um so like how do we get to like on every stack on every piece that you're not locked in there's always the uh kind of an alternative and it's no cost to switch to that alternative I think like again consensus is a really good example of that is like it doesn't matter if one of the nodes goes away because like the the switch overall yeah is nothing so it's not just the fact that you have the oh you have an alternative option is that it's also easy enough to be available at the same time for somebody to switch over yeah because like you have alternative Banks but switching from One bank to another is painted yes right like you have a turn and like if something happens with your bank like it's crude so like here you know if you have switched switches ever like kind of low switch cost no switch costs everywhere it also means that everyone is competing as hard as possible for your business right and which means like you have to derives perfect competition which means like actually everything is yeah absolutely well Ilia this has been a wonderful chat thank you so much for taking the time thank you and can't wait to see more BOS on the world