This Is The Biggest Opportunity in AI - CEO Illia Polosukhin

Transcript

avoid the collapse into effectively 1984 where you have a big brother watching you at all times and really feeding you what you should see and what you should uh do. >> Blockchains are important to solve some of these issues with bias and censorship and privacy and all these things. >> Should we be able to educate, persuade, and kind of empower people? Welcome back to the room. >> This is your home to learn from the biggest guests, the brightest minds at the tip of innovation. Our goal is to help spread the collective knowledge of everyone in our community. >> We are breaking crypto out of its echo chamber and taking it to the real world.

All right, welcome to episode one of AI super cycle presented by Near. Today we have the near CEO co-founder Ilia Pleskin to talk about the near vision for AI and really just uncover and unveil this next generation forwardthinking model of AI. Ilia, it's great to have you on. Uh it's not your first time uh but it is your first time on this new show. So welcome on in. >> Thank you for having me. excited for this uh show overall and excited to dive in.

>> I think we're pretty excited about userowned AI and uh some of the things that you have cooking over at Near. And so, you know, we call this show the AI super cycle. Uh I guess let's just get warmed up here. Are we in an AI super cycle Ilia? >> I mean, I call it AI last cycle because I think uh everything from now on is going to be done by AI. But I mean for sure you know if we think of kind of broad technologies that have reshaped the world electricity uh cars internet uh I'm you know AI by its magnitude is probably going to be bigger than those and so it obviously builds on them but I

think kind of bringing intelligence and and kind of allowing to program it to define how it's done to also leverage it ourselves. I think it will reshape the world in massive massive ways. >> And I think just the the the spending right now in terms of AI investment is certainly in a super cycle. I I think you could argue it is propping up the entire stock market in the US currently. >> For sure. Yeah. I mean I think like we definitely see the infrastructure buildout phase, right, which we've seen kind of before with internet. Um but at the same I mean and like there is discussions this is this is a bubble or or not a bubble but at the end I think

the opportunity that it brings like similarly how even though there was a lot of investment and potentially bubble that happened in you know end of 90s in internet right we know now that internet was indeed massive you know shift in how we do work how we consume information how the world operates and so uh kind of AI is doing the same but I think on accelerated terms and so that's why we kind of see uh you know a lot more um investment a lot more kind of hardware being built out because uh it's very very energy and uh kind of comput consuming

>> yeah I'm I'm thinking about kind of the power laws at play here now like you know there there's tons of investment we saw you know uh OpenAI and Oracle did this $300 billion deal recently. Neither, you know, the the joke is that neither Oracle has has $300 billion to give and and Open AI doesn't have three billion 300 billion in in uh compute to give. So maybe before we get into some of the nitty-gritty around the technicals of of what Near is building when when you see some of these massive partnership announcements, but underneath the surface h how are those deals done? what is really going on underneath the surface to to allow them to make a deal like this?

>> Yeah, I think I mean to to draw the line with crypto, right? Um you know, crypto thought they invented the partnership game where you know you partner and like both tokens pump. uh but you know the we're we're uh definitely not the first at uh kind of these games. So traditionally like in in you know in Treadfi there's a lot of deals which are done on on leverage and on uh uh effectively loans and especially when we're talking about this capex heavy work. Um so Stargate example right Stargate example was saying hey we're going to have 500 billion of hardware

built right 500 billion project of building hardware now the thing is like if you want to purchase for example I mean let's use small numbers$und00 million worth of compute you don't actually need to pay out 100 million from your pocket you can say hey I'm going to put in 20 million of from my own uh pocket and then I'll go in and borrow 80 more million, right, to buy a 100 million worth of compute. And so this is very similar how you do with house, right? You have a 20% down payment and then you know you pay out x amount uh every month for the mortgage uh except here the compute actually

makes money, right? So when you get 100 million worth of compute, you're actually making revenue from that. So you can pay back the loan while also uh taking some revenue. And so now you just did a 5x leverage effectively because you know you effectively put in 20 million of your own money while leveraging up. Now the people who are playing these games with you know 500 billion obviously are way better at this. Uh and so their their leverage is is higher. They're, you know, when they announce 500 billion, they effectively split it, you know, as an investment over five years. So this is already 100 billion a year. And then they also say, well, first year is not going to be 100 billion. We're going to only put 20

billion first year into the ground, right? And then scale up from there. And so the 20 billion, you only then need, you know, five billion of actually your money maybe or even four. And so like it all kind of goes down to uh like the exact implementation of that becomes like a financial engineering problem. And obviously like there's really good people there designing this. And now you know if we're saying this year it's four or five billion. Well, uh, OpenAI is already making almost 20 billion in revenue this year, right? And so like now this is actually a manageable numbers. Uh, and given their growth has been such parabolic and they project

that growth as well, you know, they can expect that next year they'll be able to put in more. So like it's it's all kind of I mean I think the interesting thing about AI and where kind of it it interestingly overlaps with crypto is that a lot of AI right now is financial engineering >> right it's all about like how where to get compute how to get the best like best price for electricity how to you know combine it there's a lot of actual electrical engineering obviously and and software engineering that comes in but on the surface like this big numbers are all financial engineering and where again crypto it's potenti potentially has you know a huge role to plays because you know we have instruments to do this kind of things uh although I

don't think right now crypto has actually the depth of for example you know you can't go and and borrow 20 billion uh of of capital on chain uh but I think the opportunity is there to really kind of create new instruments and new assets uh for this kind of stuff >> yeah and I think that the this deals really just signifies that OpenAI is going for you know to be one of the larger companies in the world. I mean whether public or private or you know eventually going from private to public. They're signaling like look like you know this is something that uh we're basically our product's going to be used by everyone worldwide globally everyone's going to be subscriber you know we're going to you know reach that

level of Google apple kind of uh you knowesque type of capability which is extremely promising for the entire AI space as a as as a whole but I think now is a really good time to kind of segue into some of the things that you guys feel from the near perspective is is extremely uh pressing in today's time. So you know you being one of the co-creators of the of the transform model um have you know outwardly suggested many many times on our show and other shows that the current centralized AI systems are unsustainable and frankly fundamentally broken. um you know you've been a big advocate for userowned AI which emphasizes the need for privacy and confidentiality I think

in you know prompting and data sourcing in uh the training of different LLMs and basically what we're trying to do here is unlock the full potential of AI in a per in a personal and a professional context um and so you know broadly when you're looking at this vision Ilia you know we're trying to prevent a future where a single AI can control in information. We're trying to create a more peer-to-peer network in the state of, you know, Byzantine fault tolerant networks like blockchains, but for AI systems. Um, and basically we're trying to create individual AI agents that can interact directly and autonomously in the system. So there's a lot to kind of uh overview here, but I would like to

just hear it from you for as the first episode of the of this show to really set the through line for the message that we are going to carry uh on your behalf as we go and grow this uh show. you know, kind of what is this core message that you have to the AI community, to the world, frankly, for how you would like to see the development of these systems go uh for the better of humanity. >> Yeah, I think it's a good segue from saying, hey, we expect AI and you know, OpenAI is trying to play this role to be this kind of general available technology that everybody in the world uses, right? Again kind of I mean to draw the parallel with internet right

you five billion people are right now connected to internet that's how they consume information that's how they connect to each other that's how they you know shop they and uh kind of entertain themselves and work and so as AI becomes that as well um it's extremely important that we don't end up in AOL world right where you know imagine AOL was the only company that provided internet globally to everyone and you know you want to host a website you need to go to AOL you need you want to search for something you know AOL is effectively only option to you know sees the results right that's a very you know scary and kind of dystopian world and you know very

very kind of luckily we haven't end up in that right and kind of the uh openness of the internet was uh preserved and so similarly kind of how do we ensure this happens with AI and again I actually believe AI is a more sh like world changing technologies than internet even is because internet is still kind of allows us to go and you know see the information and kind of decide if we want if we like it or not if we believe it or not etc with AI what's going to happen and kind of you know looking into the future

the AI will effectively be our lens into the world, right? You know, be that very explicitly, right, where you're wearing the glasses and you know, there's an AR on top or implicitly where you know, as you read in the news, it's all generated by AI. Uh, as you know, then like the messages kind of all the content, all the consumption, all the decisions that you make, right, they're going to be either done by AI or augmented by AI. And so whoever controls that AI, whichever kind of decisions, you know, and biases were introduced in that AI system will decide on how people are actually perceived the reality, right?

And so I think that's a very kind of scary situation if that is effectively controlled by one company, however, you know, however good or bad that company is, it's just like that company always will have a singular bias. And so that's where kind of the idea of user on the AI is not actually like open source or closed source. Um it's certainly about AI being yours, not theirs, right? Like how do we make sure that everyone has, you know, right to AI in a way and uh make sure that everyone has control and kind of sovereignty over their AI uh and kind of avoid the collapse into effectively 1984 where you have a big brother watching you at all times and really feeding you what you should see

and what you should uh do. So that's kind of the fundamental uh and I think it's very much born out of kind of crypto values as well. Um but it's way bigger than you know kind of crypto uh have been trying to do especially in the past couple years right where it's been like hey this is a new financial system to really like hey this is a new like internet right kind of back back to those discussions we've had you know in 17 18 19 when we were like hey we actually want to rebuild internet in a new way uh and I think now it's pretty clear that inter AI is rebuilding internet right and going to re change how we interact with internet and we need to make sure that it's built on a

kind of the right ways and so where this kind of goes to blockchain is I think you can't expect to try you know like let's imagine somebody right now says like hey we'll do you know your AI but we'll do it as a for-profit company I think the challenge here is that you're going to end up in a very similar situation as like the existing companies have end up because it is very resource intensive. It requires massive amount of capital and financial engineering to really scale it and bring it to the world and you need to raise capital to do that. You need to effectively do all of those actions. And so if you are doing it in kind of traditional

for-profit way, you will end up kind of still o owning the control and governance of the system. And so that's where we need a decentralized right at at the fundamental level system where you can both raise capital in decentralized way and kind of distribute value to the users themselves. You can have people participating. You can have people kind of forking and you know making it their own routes where they need to to really kind of create this ecosystem that is uh building user on AI. I think you make a really important point about AI taking the place of the internet as the lens in which we see the

world because you could say that for a long time Google you would be kind of your gateway. You would search and then that would kind of open up the encyclopedia of knowledge that that takes place on the internet. Now you could argue it's social media. AI is replacing search. Also social media companies are integrating AI deeper and deeper into their systems. Meta obviously with Instagram, Twitter has Grock, each of these systems are now integrating AI. And so I think we all have this this sneaking intuition that blockchain is a solution for some of the problems that it could introduce. But could you more just could you just go a little bit deeper into why it's a

problem that these are the companies and the centralized nature of these systems is are the ones that are pushing AI like why is it such a problem that these each of these companies has biases and they're implementing it why why could that turn out so dystopian >> yeah so maybe it's worth starting with like AI is not a new thing right obviously we have LLMs Jared of AI uh kind of popularized by Chad GPT but the thing is Google has been doing AI since effectively founding right they largi PhDs and the search the page rank is the AI algorithm of 90s uh similarly the feed that Facebook built right has been

using kind of machine learning AI principles so the reality is like we already seeing internet through AI lens for decades now and we know all the challenges that that brought right like we have constant you know battles around have this been censored or are the content being shown in like uh right ways there is like pressure there was pressure on you know social media to remove some content or censor some people and then there's another pressure which is when they don't censor like is this the right approach or not so we already have all this kind of issues uh like for the companies that have you billion user scale and AI is just going

to exaggerate that right I mean we like actually past couple weeks been interesting as a confirmation for this right we've had you know entropic and others just change their policy to start training on your data we have uh chat GPT effectively saying hey we actually running analysis on all your chats and sending them to humans to look at it and sending them to police if there's something flagged right So um >> what >> we have uh obviously Sam saying that not nothing you say to Chad GBT is privileged in any way right I mean they're already sending it to police but also like this can be subpoenaed at any time uh and so so like we don't even

need to go into like hey you know there's a malicious act of you know one of these companies like I'm I'm not even talking about that I'm talking about like very practical things where as these companies become big they effectively succumb under the whatever like political and kind of uh societal pressures are and there's no right answer. We know that like that's why block that's why blockchain is such a >> again the neutrality is really important right because you cannot be kind of sus susceptible to any of those pressures because as soon as you give up one you start giving up the other. the, you know, these are the same companies that succumb to the censorship uh regime that was taking place during, you know, some

of the elections in the United States. And there were several, you know, >> controversies that ended up not really turning out to be controversies because these these social media companies really did succumb to the censorship pressures that that you're describing. And ultimately, we'll get to blockchains and why they're censorship resistant and how it helps, but yeah, please finish your thought. >> Yeah. Yeah. I mean, just just to give you like a bunch more examples, right? So if we go to a more malicious approach is well there's this concept of sleeper agents. So you can actually train the model in such a way that it doesn't uh behave abnormally when you test it benchmark it or in normal ways but it actually changes its behavior at a specific moment of time during specific

conditions. So if we imagine how the next stax net if you're familiar this hack that happened with like Iranian uh centrifuge happen like the next stax net will be just a sleeper agent inside the coding LLMs where as somebody you know write some code like it would activate and effectively inject malicious code into the their codebase and the thing is like if you don't know what is in the model you will never know um like that that this happened uh we have you obviously like uh kind of unclear where where to stop, right? Um on on the like I mean there's this complaints that uh entropics quality of

the model declines during daytime, right? >> So uh you don't know again which model you're using. You don't know what quality you're getting at this moment. Um so kind of is this the you know the best model you have? is this what you paid for etc. So like there there's already a number of challenges even before we go into but but the the really scary ones is like right now if you go in and prompt you know take one of these models and prompt it like hey suddenly change my opinion about this topic right and just put that into system prompt on your uh in your chat GPT or whatever uh and it will start doing this across all the chats you're doing right it's

effectively in system prompt and it's tell like as you're talking about something else it will keep keep changing your opinion and so like this is totally like you don't even know if this is happening right this can be in the current system prompt right now like everybody should like like you know whatever CEO of whatever company right um so so it's really like again important that we know what goes in because otherwise we effectively giving up the control of the information we are consuming and again people we already see from the you know empirically people are trusting more and more this AI systems especially around things like you know like talking to them as therapists talking like you know making decisions etc and I think actually

that's going to accelerate because I do think AI will become our kind of delegate of decisions in the world and you know be that political decisions right who to vote for uh who to vote for what to purchase etc you know which car to buy et like all of those things will you know we'll be using AI to kind of make decisions And uh yeah, we have no if we don't know what goes in, we don't know what you know what's in the model, what's in the system prompt, you have no idea kind of how you've been impacted by any kind of biases.

>> Introducing extended exclusively on Stark, a nextgen perspit vault for attractive returns. XMV compatible and engage in unified margin spot trading, lending and per trading. >> Are are the system prompts for the major models an anthropic, claude, openai, chatbt like are are those currently available? Can you can we go find the system prompts for for these?

>> So pe people have uh so there's a ways to like you know kind of trick the model to output it but again we don't know if this is the real system prompt or they told to output that. M >> but also even if it's not system prompt you can just train the model like effectively whatever you go in system prompt there's ways to like kind of pre-enccode it into the weights as well so like there's very hard to kind of extract back the like this kind of if people really want to hide it it will be really hard to extract that kind of bias like you need to know the the pipeline what goes in >> got it okay so now I think this sets up really nicely for a credibly neutral censorship resistance

base layer that is totally verifiable so that you can see the system prompt, you understand what goes in and then you can understand if it's being tampered with. And so we can see not just the input the original input system prompt but we can also see all of the potential attack vectors and the ways in which we're injecting bias into the system and and this at at a theor theoretical level is blockchain but I'm curious more about mirror's implementation of this. I think you were starting to get into why blockchains are important to solve some of these issues with bias and censorship and privacy and all these things. Maybe you can give a little bit more of a

general, you know, solution as to why blockchain is the answer, but more directly what why is near blockchain the answer for AI's problems? >> Yeah. So, I would say specifically blockchain is component of the answer, right? So you need like a whole system um to to really deliver a a result here. And so what Near combines right, it's not just blockchain. It combines a set of technologies. Uh specifically we have a technology called decentralized confidential machine learning. So this builds using blockchain a decentralized uh kind of network of accelerated hardware that runs in so-called uh secure enclave or trusted execution

environment mode where all the compute that happens there is confidential private and verifiable um and and it's a very pragmatic choice because all the cryptographic primitives right you know people love ZK people love fully homework encryption they're too slow they're too expensive and they're not practical uh for use in kind of AI workloads. And so we have this kind of pragmatic approach where you you know give up some of the sovereignty given the kind of the hardware uh manufacturers involved but in turn get you know this verifiability and privacy for it's it's less than 5% overhead usually 1% overhead uh to the normal compute cost. So, so that gives you

effective ability to run AI agents as well AI inference and AI agents in a verifiable and confidential way. So now you actually have a guarantee that this specific model was run on and you I'm as a user know that it run was on my data there's nothing else ran on my data and this is outcome and nobody else will be able to get this data and this this run. So that's a really important component. It uses blockchain for uh you know the marketplace itself for compute for the payments for uh kind of uh coordination of this whole system as well as key provisioning and management. Uh there's another component which we call chain signatures which is a ter validators themselves run multi-party computation

network to allow for data encryption decryption as well as uh kind of signatures on for other transactions on other chains. So that allows to have actually end to end encryption for the data. For example, you can encrypt your private uh information, memory, you know, connections to, you know, Gmail, etc. And it's only decryptable again inside the agent that runs on your behalf. Uh and then that is also allows us to run autonomous agents and kind of other use cases. So it's really like a combination of technologies that working together uh to provide us with a solution. But indeed solution is uh kind of from a user perspective offers you

effectively a private AI that you can use that you can trust to add a lot more context right like one of the challenges like I'm I'm you know feeling feeling very unsettled to try to connect my Gmail for example to open AAI right because my Gmail has a lot of you know websites and like uh access controls through it etc so like if something happens like who who knows like you effectively can be compromised across a lot of places. Uh and so you really want like to know that this data is not going anywhere that this really uh kind of stays private and similarly in a worker environment the same thing right like you know you guys running the podcast

you you definitely don't want all your data to like randomly go somewhere um and especially financial institutions legal healthcare like there's a lot of industries who for whom data is extremely sensitive so this opens up an ability to really use all your data all your context in AI uh while knowing that effectively it's as if it was running locally right uh but really getting the best of the best models uh open weights models now further than that it allows us to start actually building fine-tuning and then building models in a new way where we actually know that this data was run you know with this code and this is the model we received

and so that's kind of the the new thing we're introducing we call it open uh like open- source but encrypted weights models where you know what goes in but the model weights themselves are stay encrypted and so because of this you actually can monetize them. So you can have now a monetization loop that allows to fund research in kind of and development of this model including compute and in turn receive a model that can pay back which means also you can now you know do financial engineering. So you can actually have a token per model right which you can distribute you know as an investment uh for people who contributed to it and then the token will capture the revenue from this model

going forward. So that actually allows us to to bring you know this kind of verifiable understanding of what what goes into the model and because it's an open platform you know if you don't like what went into into that model you can go fork it you know train your own version with maybe different biases and use it but at least now you have this choice and you kind of have a monetization framework around it to also benefit the contributors and developers and and comput providers. >> Yeah. So, so I I would say then like the core the core idea here is around the user data the memory the context and models slashinferencing being private and owned by the user. This is like the

core idea that we must always get across. >> Correct. Yes. And then basically this we we we can achieve this through NAR's you know setup of effectively what I would consider like a a decentralized confidential like cloud-like system where uh you know you have GPUs and CPUs that run in these kind of secure confidential vaults and then you've got you know on top of that you have things like chain signatures and you've got the uh you know in uh encrypted model weights that that you just spoke on and so effectively we can create this end to and encrypted system uh where models can only be decrypted wi within these like confidential uh you know like scope of

scope of of agreement basically saying okay like we we are going to allow the uh unencryption or the decryption sorry of these given model weights but only in a secure environment and basically what what we get to at a at a point as a result of all this should it work should we be able to educate uh you know persuade and kind of empower people. We should end up getting a world where you feel comfortable connecting your Gmail and your YouTube and your and your Google Docs and your Google Drive and everything to your to your LM. And that will give you a much better experience, much better result. Uh but also maintaining the same levels of core privacy that you know that you would

expect from any other kind of large tech company or just in general uh kind of like what you're doing on on on chain. And so the the future could look like a better version of what we have today because users are willing to provide more information to be more open about their personal life but also just connect other APIs, connect other services into one um and into this kind of fully private userowned setup. >> Exactly. Yeah. like it the the idea is that you can actually get a better product because you are able to connect more context and again right now AI I think the AI capabilities are you know really good it's really now about

context and in turn you know going forward it creates us a model where we can actually know what you know we get the organic mo organic you know far farm uh farm-totable model where we know what goes in into this as well. Uh versus right now, you know, you you have no idea what kind of GMO is in it. >> Yeah. Things are slammed full of antibiotics putting putting microlastics in our AI. Um IA I think you know people have if you've been around the space and and you know you've probably seen near uh because you guys have been around a long time, you guys have survived multiple cycles. um but you have um built pieces of this technology as you've as you've

gone and progressed towards this vision and so I can remember when you know near was you know it was an L1 kind of and then it was chain abstraction and then now it's AI and I think to someone who hasn't been paying attention this might feel a little bit discontinuous so I just want to give you a chance to explain the continuity behind this vision And then ultimately once we've kind of gotten, you know, up to speed, you know, where where do we go from here with near and and how it is intertwined with AI? Could you just kind of could you just kind of like tell the story and catch people up and then and then try to forecast out and tell tell what is coming up next for Near?

>> For sure. Yeah. I mean the story we started as AI startup right back in 2017 effectively after I worked on Transformers. I left Google because I wanted to put that into production and and build a product and the realization was it was too early like we didn't have enough compute. So what we were trying to do was to get a lot more training data. Uh and so we were coordinating a lot of uh computer science students around the world to get us this training data. And we were lacking a global payment network that we can use to actually pay them right like into China into Eastern Europe into Southeast Asia.

There's there's all kind of challenges paying in in the fiat world and especially when we're sending like 15 cents, 30 cents here and there, right? uh neither Ethereum or Bitcoin were able to do that in 2018, right? The fees were higher uh even back then. And so, Near Protocol started as like, hey, how do we actually build a network that is able to kind of coordinate this kind of uh you know, data contribution, data labeling type thing. But obviously, you know, we went and talked with more developers and realized that like everybody has this problem, right? You know, not just not just for machine learning. And so uh kind of we ended up building this generic layer one because we wanted to make sure everyone can use it. And so uh

interestingly the data labeling platform actually launched in 2021 and it's been running ever since called uh kind of near crowd. There's also a few other uh data platforms like public AI that launched uh since then. But the as part of this vision back then we kind of realized like you know the vision of AI like near AI originally was kind of changing how you interact with computing because if if your AI can write code I mean kind of what we discussed right the way you will interact like you will not need apps you'll need you will not need SAS you can just have AI build it on the fly and then we bring the blockchain we realize like hey actually you can also

change how the data is done and how the ownership is done. And so back in 2019, if you go back and actually listen to some of my talks, we're talking about, you know, user on internet, right? And I would say like, hey, we need to replace Google and Facebook and build user on internet. And the problem there back then it wasn't clear how to do that, right? Everybody's like, well, how how you going to build a decentralized Google? That was the question I would get. And I'm like, well, you know, we need to build the foundational blockchain technology and then we'll, you know, build stuff on top. Now, with LLMs now, it's pretty clear how to build decentralized Google, right? I don't think anybody is actually asking this question anymore. Uh, even you know, social media as well, like we have we

have answers now. So that was always kind of like hey the transformation is going to happen uh and it's going to happen kind of across the whole stack and you need blockchain as a way to you know um around it kind of root of trust coordination microp payments etc. Now with chain abstraction again the part of the realization was like well as we're building this it was actually clear that you know there not going to be a single chain that wins everything right there going to be applications everywhere they're going to be kind of smart contracts and value everywhere and so you want to unite it and again AI will be the interface but you want a way to really give AI like a way to navigate the whole web 3 and so chain abstraction

kind of for example chain signatures were built to navigate the web 3. It's also used for encryption, decryption of the data for the AI, right? So, it kind of serves multiple purposes. Um, uh, I had this discussion as um, uh, uh, Justin Bonds actually like he's like, hey, wouldn't AI just go and like have accounts on every chain? And my answer is like it can, but then you're losing all of the kind of controls and uh, kind of systems that you can attach. And so if your AI is already running on near in verifiable way, well your near account now is kind of the control plane, right? Where you can say, hey, you cannot do this type of action. You cannot kind of escape this. You have

near intents that allow to execute actions across all of these chains. And you have kind of near chain signatures to really kind of control the assets and and value across all of the uh different platforms. So it's really like kind of composes all together into one stack. Now going forward I think I mean we're definitely continue you know building out this TCML kind of this private confidential and verifiable cloud part of you know use cases you can use it for is to run for example autonomous agents right where you run instead of smart contract you run the AI agent in this kind of confidential verifiable way it you know can run on any of the nodes available on the network so you cannot

shut it down you can give with a mission you it may or may not have token holders and now it can go and do things and again intents serve as effectively a way to execute actions in the world right be that you know onchain or offchain so kind of the way I I assume like you know and we're kind of building toward the vision is every individual has their AI that is really their interface to the world right their lens to the world it is able to go and execute actions both onchain and offchain and using intense type architecture, right? So everything from you know give me some Bitcoin to go and uh you know buy some pizza or you know I want to build a you know house

for my for my kid. Um and then on the other side so this is like individuals and they kind of using the intents between each other as well as with other autonomous agents right that are running in this they have a mission they have maybe have uh kind of shareholders and so that all together this whole system effectively is this like super intelligent system where you kind of have lots of this AI together you know maybe solving problems together as well uh and so that's kind of the vision we're building toward kind of lots infrastructure privacy user ownership but also kind of how all of this forms you know a super intelligent swarm as well. Um but all of this you know we continued learning and iterating right

because like uh as you mentioned right we we we kind of focused on different angles of this vision at a at a time because you can't like you know ship everything at a time but as ecosystem expands we actually have different teams going in all the directions now in parallel and so uh there's definitely still teams on near that are building ton of innovative you know D5 primitives that are multi-chain uh there's teams doing AI that kind of leveraging this. So kind of the way we think of this is like this is going to be a multiaceted ecosystem that together forms this kind of really innovative way you're going to interact with with new internet

>> and Ilia I think that that you can tease maybe a little bit as to one of the upcoming products that will launch uh this month or early next month for near AI or at least kind of a uh maybe give us a little bit of a hint. Yeah, I mean so there is a fully private end to end encrypted chat. So that you know should replace your chat GPT and let you uh kind of day-to-day switch uh to the product. So we've been we've been kind of testing it with few of the close partners. Um and yeah, pretty pretty excited for this uh coming to market because this this is kind of first version of really like all of those pieces working together and showcasing how to do this. It also serves as

developer platform because like as a you know everything you you as a user you use it if you get go and log in later in other applications with the same account those applications you know on your permission can access your context as well. So this means that new applications can sprung up kind of leveraging this and kind of keep expanding the the network. So I think there there's a lot of exciting stuff coming up and uh yeah I'm looking forward >> and that'll operate in that same like you know encrypted and then decrypted you know based on user user uh preference but w within a controlled environment kind of setup correct >> correct yeah >> beautiful Ilia Paloskin episode one AI

super cycle one of the co-creators of transformers at Google now with Near AI we have a this is like the beginning of a parabolic S-curve rise on this show. Ilia, we appreciate you being the first guest on a Sunday. Um, and for helping us put this all together. Robbie and I are on the road to Jensen Hong on this show. Um, and uh, we're not stopping until we get there. So, appreciate your time and uh looking forward to to seeing this new chat product come out, speaking to you again in a couple of months as well as just you know co-hosting some of these together if we get a really awesome guest that you want to be a part of that you think can uh be a part of

the near ecosystem. So, Ilia, appreciate your time, Ed. >> Sounds good. I appreciate uh you guys doing this and excited for what's coming. >> Beautiful. Thanks, Ilia. All right.