Brad Keoun and January Jones dig into decentralized AI infrastructure with a news roundup with Evin McMullen from Billions Network on agent payments and identity, then two field interviews from Stanford: Michael Heinrich (0G Labs) on verifiable, privacy-preserving AI, and Illia Polosukhin (NEAR Protocol) on staking for AI inference and NEAR's new agent marketplace.
The news roundup with Evin McMullen covers two DeAI News stories: a new Shoal Research report on why USDC's dominance of agent-to-agent payments on x402 is more fragile than it looks (Coinbase controls the clearing layer, Stripe and Tether are building alternatives), and Cloudflare's launch of stablecoin wallets and identities for AI agents. Evin connects both back to the same problem — agent payments only work at scale with an interoperable identity layer — and brings in the Air Canada chatbot ruling and a conversation with SEC Commissioner Hester Pierce on agent accountability.
From Stanford: Michael Heinrich explains 0G's case for open-source AI on safety grounds, how trusted execution environments make AI computation verifiable on-chain, and 0G's $350M raised to date. Illia Polosukhin walks through NEAR's new staking-for-inference product, confidential AI via secure enclaves, and why NEAR just joined the x402 Foundation.
Guests
Evin McMullen, Co-Founder & CEO, Billions Network — http://billions.network · @provenauthority · @billions_ntwk
Michael Heinrich, Co-Founder & CEO, 0G Labs — @michaelh_0g · @0G_labs
Illia Polosukhin, Co-Founder, NEAR Protocol — @ilblackdragon · @NEARProtocol
Timestamps
0:00 – Cold open
1:17 – Welcome to DeAI News Journal
9:50 – Today's lineup
10:20 – News Roundup with Evin McMullen (Billions Network)
10:44 – Story: USDC's lead in agent payments faces new competition
17:53 – Story: Cloudflare launches stablecoin wallets and identities for AI agents
35:53 – Feature interviews from Stanford
37:39 – Michael Heinrich, CEO, 0G Labs
56:12 – Illia Polosukhin, Co-Founder, NEAR Protocol
1:14:06 – What's next + sign-off
More from DeAI News:
DeAI News covers the infrastructure, cryptography, and protocols defining sovereign AI — decentralized compute, agent networks, and crypto payments for the machine economy.
Read the stories and catch new episodes at http://deainews.com.
A Distro Media Production
Produced, edited and designed by January Jones
Music licensed through Soundstripe.
Code: 7IIILTUES54J7TXP
Transcript (by Riverside)
January: All right, we ready?
Brad: Mi va
January: VO bless us in our endeavors. May you come back from your outage.
Brad: May you recover
January: Ha ha ha
Brad: to be something even Possibly starting to come close to Hermes. I didn't say that. That's so mean.
January: May you decentralize your infrastructure so not every AI model is out at once.
Brad: May you find the stack you deserve.
January: that's a good one. All right. Well, we opened our show with blessings. Welcome to the DEAI News Journal. Brad, we were just talking about hearing the news about an outage this morning. And that's how we got into our blessings for AI. that multiple models were out at the same time. And so This goes to kind of our point and why we're covering decentralized AI on the show because this brings up so many questions from people, like why all at the same time? Aren't those different models different companies? Well, maybe they're running on the same infrastructure. So we can dig into that later, So this is our second show, and we have been hitting some themes as we launch generally. So the first show we tried to kind of lay out what is DEAI, what are people saying about it, and and kind of show you the stakes. I think the Shaw Walters interview did a good job of kind of going deep into tech, but then also saying, we're talking about access to education. We could be creating an underclass since that interview, South Korea announced AI for everybody, right? No barriers to a model as a country. and that relates to what we're doing today. So today we're talking to founders who are leveraging blockchain for the new AI era. And so we have Evin McMullen from Billions Network, and she's gonna do the news roundup with us.
Brad: And then we will feature a couple interviews we did while at Stanford and Palo Alto with Michael Heinrich of Zero G Labs and Ilya Polosukin, co-founder of Near Protocol.
January: And we are still using all the content that we did gather in San Francisco, but we have decided this is where we're planting our flag. So, Brad, you're in Austin. I was previously in Denver. I came out to the Bay Area because this is this intersection, and really the people that are innovating and are thinking differently. and so that's what we're trying to connect to here. Austin itself is also. such a tech hub and you're getting out to meet people and there are lots of founders in the space in Austin as well. So we feel like that's a really good balance, Austin and and San Francisco. before we get into the segments today, we have been defining ourselves for DEAI news in the open. We have this build-in-the-open idea that we do with distro media where we we publish DEAI news. but while we're doing it, we're trying to define what we are as a news organization, and be very conscious of what we can cover today and where we're going, right? So we are bootstrapped. but I just wanted to have a small discussion today about like what is DEAI news right now, and what can people expect from our coverage, especially on the journal.
Brad: DEI News was launched late last year, initially a telegram channel, and then as we built distro, we launched it as its own publication on distro and now we're expanding the whole franchise to include a podcast and basically it's just us and I'm d going through the DEAI news. I'm using a few tools that we've built to kind of monitor the news and then I'm using distro to publish DEAI news. So right now I would say DEAI news is probably think of it more like a trade publication in the very early stages. it's probably a more of a B2B kind of focus. We're we're really it's it's pretty hardcore, you know, tech news for a pretty hardcore tech audience at this point. We would love to build that out. I think probably the main value that I'm bringing right now is the filtering of the noise I am curating or or producing. Only one or two or three stories per day on DEAI news I think just being realistic, I mean we want to be fair and independent. We don't have a ton of time right now to do investigative deep dives or huge like data driven journalism, And then here on the show we talk about the news, but then we also kinda have a chance to bring out some of those human insights too, right?
January: Yeah, I think we're in the phase of launching and expanding the brand and we are very much in an explainer mode where we are
Brad: Yeah.
January: needing to explain. From my background being in public radio, there was a lot of rules, no calls to action, you can't s sound like you're selling a product, right? And so I think
Brad: Yeah, yeah.
January: the editorial perspective, you know, when I'm reading things or I'm talking about how do we even introduce our guests on the show without sounding like we're rereading their press release, right? And so it's like this this balance of giving enough context and information about what these people are doing. But not going overly into this turning into a a product podcast or it sounds like we're endorsing or we like their product in some way, right? And so I think that's part of my editorial balance when I'm thinking about What is news? What you know what is newsworthy? Is it that there's a new technology and here's how it works? And I think that's where we are right now in the phase of of DEAI news because it is something that we have to explain again and again, really with every story and introduction, because it takes a while to stick, even with us. We end up spending so much time researching, we have no time to record sometimes. cause we're
Brad: Well
January: like, wow, what is that thing? What is it doing? but that's the journey we're taking everybody on in the journal, right? Because we're not experts on all this. We're just like everybody else. Trying to understand how it fits into the technology that we're using, but also into the power systems that we're seeing. Because it's a very big deal that these AI companies and their models are running policy, running access, like we were just talking about. They all go out for a few hours. that is an extremely alarming volume. vulnerability, right? And that's all in the hands of private companies.
Brad: Yeah, I I mean I I would say that i if we're right in our hunch, which is where a lot of journalism starts from, right? It's a hunch. And then you start following the story. But our hunch is that decentralized AI, as we're defining it very broadly, is could be the foundation for the entire global economy. I mean that that's a lot of some of the projects we're gonna talk to today, they s see the An agentic economy being built on these kinds of projects. And I mean, we're not saying who's going to win, but I think we're saying that, yeah, these some of these builders are tackling the problems that need to be tackled in order for our civilization to survive this AI revolution, right? And I I think to your point, some of it is very, very technical. It's so technical. And so we're A little bit like science writers we were talking about the other day, where we're, you know, our job is really just to sort of explain it so people can understand it.
January: Well, that is us providing this translation layer the best we can, but we're balancing that by letting the people who are building these products explain it themselves, right? They've got the best analogies, they know how to to kind of translate what they're doing. And so let's get into that. So first we're gonna
Brad: Yeah.
January: do our news segment. You and I talked to Evin McMullen from Billions. Billions Network uses cryptography, zero knowledge proofs specifically, and it's anchored on blockchain and it's giving verifiable human identity for AI agents. So this is the theme through our whole show today. We're talking about builders looking ahead and building for an agentic economy that's almost here, but not quite yet.
January Jones: welcome to the DE AI News Roundup. Today we have Evan McMullen, co-founder and CEO of Billions, an identity verification network for humans and AI agents. Thanks for being here.
Evin | Billions Network: Absolute pleasure. Thank guys so much for inviting me.
January Jones: Well, it's good to have an expert in the house. We're gonna talk about some of our recent stories that we published on deainews.com. So let's get into
Evin | Billions Network: Certainly.
January Jones: the news. So we published a piece about USDC dominating agent payments. So USDC, the US digital currency of Circle, is currently settling 99% of all agent-to-agent payments. On X402, which is the protocol that a lot of people are using right now. But a new report from Shoal Research says that dominance is pretty fragile. Coinbase controls the clearing layer. Stripe is building alternatives. Tether also has a stable coin. So why does it matter what stable coin agents are gonna be? using and settling transactions with. I mean we've had this new kind of world in the United States open up for stable coins with the legislation that passed last year. I mean I ask myself what stable coin should I use and why does it matter? What does this matter for agents? Evan, what's your take?
Evin | Billions Network: So historically, the goal with stable coins has been to increase deposits and liquidity on chain. And as you know, Tether and Circle have quite an early lead here. but everyone is entering the stablecoin race. so the only way to lead to a coherent outcome, a coherent experience with many stable coins in the market is if we have a clear, interoperable identity solution to determine who's an agent, who's a human, and who they're accountable to. This is especially important in the decision. Between which stable coin to utilize, because of course each has different policies around governance, especially when it comes to compliance. And we've seen the disparity in that action between, for example, Tether and And circle in the way that they handle doing things like freezing tokens in instances of illicit activity. So with everyone issuing stable coins on different infrastructure, different backends, using different messaging protocols, the really the only way that this all works together is to take care of identity such that we can identify agents, who they're accountable to, the human beings and organizations they're associated with, and then make that interoperable across different stacks.
January Jones: surprise, you wanted to talk about identity, Evan.
Brad: Evan Yeah.
January Jones: But Brad, I want to take that to you. What do you think about this? Because like, you know, you've had a front seat to the rise of stable coins and the adoption. I I really wanna know why does it matter what stable coin is being used by agents?
Brad: Well, I think it's interesting if you trace the arc of crypto and there was this race to you know, first there was Bitcoin and then there was Ethereum and then people started to launch tokens on top of Ethereum and then people came up with this idea of a dollar pegged token and that was a stable coin. and it was just a crypto thing and people were using it in payments for crypto and now In the age of AI, here come the agents, and now you have very serious payments players from the traditional world. Stripe and Visa and they're all planning to use stable coins for their agent payments. And so a and traditional commerce is gonna want to be doing agentic commerce. And so now all of a sudden the stable coins are become a much more important thing. And I guess to your point, Evan, the the sort of enterprise infrastructure is not really there yet, right?
Evin | Billions Network: It's certainly diverse, fragmented, and early in in many of its
Brad: Yeah.
Evin | Billions Network: implementations. and so, you know, I think the the the way to think about stable coins in this in this context is is machine money. They are uniquely well suited to the capabilities and requirements of agentic interaction. however, the overlap between ownership of different parts of the stack, you know, you noted earlier, Coinbase. And their you know ownership and and contributions not only to the you know stablecoin USDC, but also to payments and identity protocols like X402 and 8004. And so it doesn't surprise me that the house that built the agent stack, at least a popular one in the crypto space, is also responsible for the currency of choice of those agents. but I think we're going to see. Further customization and opinions, especially from these more institutional players, as they start to enter the chat. But this yet again underscores the importance of interoperability, of governance, and of securing the value of those stable coins, as we've seen other stable coins depeg in
January Jones: Mm-hmm.
Evin | Billions Network: the past to, you know, very negative consequences.
January Jones: Yeah, I mean, we're still spending money out of our wallet when we have a stable coin, right? And to your point, every wallet, your stable coins are associated with some kind of identity verification or, you know, your financial profile. when we're talking about mainstream adoption of stable coins, we're already like knocking out half the population. Who knows what we're talking about, right? And then we're right, they're like, What the hell's a stable coin? That doesn't sound stable.
Brad: Hahaha totally.
January Jones: That's fake internet money. You know? And now like we're drilling down and we're like, what stable coin does an agent need? And and the thing is is like, who's an agent? am I? but when we're talking about this to to bring it down a bit, when we're talking about an agent with a wallet, that could be you controlling your Claude right? To purchase and buy things. It doesn't have to be this kind of like you know, very far removed from you scenario that's not happening yet? Or is this really not happening yet and everyone's jumping the gun about how this is gonna work?
Evin | Billions Network: So I think it's happening, it's happening more quickly than we expect, but the switching cost to adopt this new way of payments and way of interaction is a little bumpy. Juniper research notes that in in the next year or so the value of the agentic economy anchored in stable coins is likely to exceed eight billion dollars, I believe. looking north of a trillion here in the next few years. so this implies quite a lot of value moving along these rails enabled by. By agentic actors. However, again, this also highlights some of the shortcomings in the supply chain between buyers and sellers, between merchants and those who they serve. Because of course, this implies participation of payment processors, of the merchants themselves, the platforms that are receiving payments. I think one of the unique and interesting challenges is how we reconcile legacy infrastructure, especially in the commerce space. with this new method of payment when questions like compliance come into play.
January Jones: Mm-hmm. Yeah, well, I think right now the US government, we can't pass the Clarity Act. Stablecoin passed and and nobody knows how to what to do with it quite yet. we don't really know how this is gonna play out with regulations yet, because it is just so new. but technology waits for no one, and AI waits for no one, And so that leads us to kind of our next story, which is about Cloudflare. So Cloudflare isn't a crypto company, but they just launched stablecoin wallets for AI agents, and this has identity handles, spending caps, in theory. No one's used. Yet, I believe. Amazon tried a similar thing with Bedrock. and so we have now those big tech companies moving into this space and showing us, I guess, how it's done. Evan, what's your take on this coming from being a founder to give identity to agents via their humans?
Evin | Billions Network: So I think this is a great indicator of progress with organizations that are well established, like Cloudflare, revisiting their definition of what a user can be in a digital environment. so
January Jones: Mm, true. Now it's agents too and people.
Evin | Billions Network: And so this bifurcates the users of of Cloudflare, you know, based experiences into human and non-human. But now we're introducing a secret third thing, a verified AI agent, thus broadening the definition of user types in Cloudflare enabled. Environments. Beyond that, it's also, I think, a strong indicator that a centralized infrastructure organization is adopting crypto native Rails rather than inventing a net new closed system. But I I think sort of lastly a a tension worth exploring here is that in this context identity disclosure is optional. And so this does not solve the problem entirely for merchants who may still face anonymous agents, concerns around compliance, accepting different types of payments And so the existing Cloudflare bot management stack that we have known for you know a number of years as as users of the internet, I think also is due for a facelift.
Brad: well I I mean I think this story is interesting. In some ways, it's the exact opposite of the circle story from kind of a a competitive race standpoint. Here, you know, with with the with the stable coin, all these almost all of agentic payments are being done with USDC. They're out to a clear early lead with this kind of key thing that somebody might want to be dominant on. And then with the wallets and the agentic identity, I feel like those are kind of areas that have not been won by anybody yet. whereas the you know the the circle story they're sort of you know it's theirs to lose now.
Evin | Billions Network: I think on the one hand, you know, there's to lose. However, the early lead that we are talking about describes very niche adoption. The experimentation of agent
Brad: Yeah.
Evin | Billions Network: payments, especially the X four two protocol, is championed by innovative developers, small organizations, you know, a a handful of institutions are are coming to the table, but much of the progress and adoption has really been driven by those at the frontier. And so I think you know, for for the the health of the agentic commerce space in the long term, for greater security and for a more robust digital economy o among agents, we are going to want to have a diversity of providers. And and that also will help to, you know, to mitigate any single vulnerability such that a single issue cannot take down an mul an entire multi-billion dollar economy. and so I think interoperability with existing legacy systems, meeting the security requirements that already exist today instead of building on the fringe is the only way that this is going to become entrenched and legitimate enough to scale.
January Jones: So I'm gonna I wanna back up one step with Evan here you've entered the space with a mission about giving identity to agents and you know, via their people, right? So I think that it that is a hard thing to tell people they need. Could you just tell me, like, in the last six months, has describing what you're doing for work gotten any easier? And are you seeing more relatable use cases where people are trying are getting like, my? Agent needs a wallet and an identity, just like my Apple Pay, let's say.
Evin | Billions Network: Certainly. I think in the in the last six months, the everyday you know, human dinner table conversation has become much more literate. but I also think that the institutional and government adoption of these technologies have started to highlight the gaps, especially around, you know, failures of of interoperable identity technology. you know, I I think anyone who's had the experience of interacting with an LLM enjoys the early surprise and delight of the experience. That is so much more enriched than something like semantic search. But very quickly, as soon as you interact, let's say, with a customer service agent or with an agent in a chat environment, becomes very easy to ask the question: how do you know that that agent is really representing that organization? Or that that agent
January Jones: Yeah.
Evin | Billions Network: is really representing Evan when you know it's claiming to represent me.
January Jones: It is. And and just for your normal user, that's just a pop-up on somebody's website and you trust that that is like a a real representation.
Evin | Billions Network: Well in and January, I think you're highlighting actually what what what alludes to some interesting legal precedent here. so
January Jones: Ha
Evin | Billions Network: last year, Air Canada, the airline, incorporated a Chat GPT wrapper, basically a little customer service pop-up window enabled by AI. And as you know, AI is wont to do, it began hallucinating plane tickets that did not exist, plane fares that were not sanctioned by the company. And so This altercation between travelers who thought they were purchasing legitimate tickets because they did so on the airline's own website. And then the airline itself, then pointing the finger downstream at the provider of that AI
January Jones: Yeah.
Evin | Billions Network: resource, ended up in court. And the outcome of that case was that Air Canada was in fact responsible for the actions of That agentically enabled, that AI enabled experience. And so what
January Jones: Mm.
Evin | Billions Network: has come out of that is the requirement of accountability for agents that the last party or the last brand to touch and present agentic information is likely responsible for that output.
January Jones: is that just Canada?
Evin | Billions Network: so that is where we have legal precedent right now that we can point
January Jones: Yeah.
Evin | Billions Network: to. But in a recent conversation earlier this year with SEC Commissioner Hester Pierce, I was asking her in her personal capacity how she would think about AI agents acting on behalf of individuals or entities. in which she
January Jones: Mm.
Evin | Billions Network: shared, and again, you know, this is in her her personal capacity, not the explicit view of the SEC, but that agents, you know, can be considered tantamount to subcontractors. That they are acting on behalf of another entity
January Jones: Yeah.
Evin | Billions Network: that is then accountable for their actions. But of course, as we know, accountability requires identity. There is no neck to choke if the neck doesn't belong to a person or an organization with a name, a mailing address, and an ability to contact them. And so This kind of concept of headless, unaccountable AI begs the question: what happens when bots do bad things? So whether
January Jones: Yes.
Evin | Billions Network: it is a question of topping up the wallet and payments needing to come from a source, or a question of diverting from policy or even transgressing legal boundaries, accountability in the agentic space is vital.
January Jones: Mm.
Evin | Billions Network: We've also seen this arise, you know, in a in a very real and tangible way when it comes to national security. there are, you know, many folks around the world employed with the access to state secrets who have begun availing themselves of the great resources of Claude and Chat GPT. But of
January Jones: Ha ha ha ha ha.
Evin | Billions Network: course, when you enter information into those environments, you are using a trust me bro security policy, meaning that you gotta cross your fingers and hope that the information disclosed. There is not going to transgress any of the boundaries, data expatriation laws, or security requirements that may already be in place. So exactly as you were noting earlier, that there is a vast chasm in between technical adoption and regulatory oversight. We are seeing that happen in real time with leaders and governments themselves.
January Jones: Yes, we are. I mean, it is fascinating because this is borderless, this is countryless, you know, agentic payments and commerce. and so how do these things play out? Canada has a precedent. How does that apply? You know, we've covered a story about the rise of this theory of agentic courts to have some sort of like place to, you know, for agents to to dispute their own issues. but this is a this is a big legal question, right? because one, we've seen the lack of ability and different rules in other countries in the European Union versus the United States, how they crack down on tech companies, big tech, Google, Meta, you know, and try to regulate them, what they think their purview is. But now we've if we're setting out all this agentic commerce. Yeah, like how do you give accountability and how do you have legal precedent and laws cross border all the way around the world and when it happens in literal seconds?
Evin | Billions Network: Certainly. And, you know, this of course is all tied back to payments. And so, you know, regulatory oversight such as OFAC in the United States, which limits the parties that are legally allowed to transact, to interact with one another, you know, i i implicitly one would think that such requirements will also apply to AI agents and their associated payments, that merchants will need to be sensitive to the source of funds and the types of parties they're they're facilitating interactions with. But right now we do not have a lot of clear guidance because the speed of of innovation and adoption is light years ahead of the antiquated regulatory framework being applied to it.
January Jones: We don't have to go so fast, right? Like,
Evin | Billions Network: You know, I mean yes and no, kinda by by the same by the same note around around Juniper research I I cited earlier. So the expectation is that agent commerce transaction value is gonna top out around eight billion dollars by the end of this year, looking like a about one and a half trillion. By 2030, but there are other estimates that by 2030, 2031, that's gonna exceed three trillion dollars. the capital is moving whether we like it or not. And so it is incumbent upon lawmakers to catch up swiftly, especially in you know, questions around, for example, taxation on these purchases. How is an agent going to pay taxes? Where does that happen? So I think there are a lot of unanswered questions that are deeply relevant to the unanswered identity interoperability challenge that you know that come down to institutional adoption and regulatory oversight.
January Jones: Yeah, well fascinating. so agents with wallets, they need an identity somehow. I'm mildly convinced about this now, for me, I have to first start thinking about it like my identity, my wallet, right? And my agent is just my smart computer helping me get things done faster, you know.
Evin | Billions Network: Yes, your personal butler, but you are still responsible for the actions of that party, or at least that's what it looks like right now.
January Jones: Brad, what do you think about this?
Brad: Well, I do think you know what is an agent i is kind of maybe where I would start the first generation of these agents as I see it is basically going to be like your computer, right? It's just one degree removed from you. you're kind of directly controlling it the way you are controlling Claude or ChatGPT. You're typing stuff in or speaking to it. and then it's doing stuff. I think there's a there are people who are spinning up twenty thousand agents and they're all autonomous and they're doing things. And I think that is probably not going to be the what most people are doing. But I don't know. what do you need privacy for, what is the agent for? I think that is just a a soup that a lot of people are just
January Jones: Yeah.
Brad: experimenting with right now,
January Jones: Evan, what do you need privacy for? What would you say people need privacy for right now
Evin | Billions Network: So I would sort of flip that question on its head. What needs to be disclosed? And so, you know, at the to sort of call call
January Jones: Okay? No
Evin | Billions Network: call back here to a reference from the the 2000s, there was a a set of design principles created by Kim Cameron at Microsoft called the Seven Laws of Identity. These are design principles that you apply if you want to minimize risk and harm to human users in a digital system where there are identities such as accounts. Now, one of those seven laws is minimum disclosure for constrained use. What minimum disclosure means is that the minimum amount of data should be shared to execute the or accomplish the particular use, right? To accomplish the goal. And
January Jones: Mm.
Evin | Billions Network: so Over-sharing data, sharing more than is required to accomplish the goal, then you know requires the individual, the participant, the holder of that account to forfeit their ownership of their own information because they do not know if they're going to benefit it from it all benefit from sharing it at all. and we've seen time and time again that the over-disclosure of data. Does not necessarily mean something immediately dangerous or harmful. But in a medium term, that data can then be utilized for purposes totally unrelated to the reason it was originally given. I think you know we we also know that in the environment where you know we have AI-enabled systems and also concern around safety for underage users in digital spaces, that things like age verification laws
January Jones: Yeah.
Evin | Billions Network: are sweeping around the world right now. you know, there are folks like Larry Ellison who would try to mislead you to believe that there are two ways forward for the internet. One where every website collects your passport, and the other one that is an unusable, you know, cesspool of bots. but I believe that there is a middle path where you can prove qualification in a way that does not over disclose data, but rather discloses only that which is required, minimum disclosure
January Jones: Mm.
Evin | Billions Network: for constrained use. In a perhaps more real-world but and lower stakes example, in the United States, if you want to enter establishments such as a bar that is age-gated, you need to share, for example, a passport or a driver's license to prove that you are over the age of 21 to enter. Now, why does the door person at that establishment need to know whether or not you're an organ donor, exactly how tall you are, how much you weigh? That is an overshare of information
January Jones: Yeah, why?
Evin | Billions Network: wildly outside of that which is required for for that particular use.
January Jones: And your home address.
Evin | Billions Network: Exactly. And your home address.
Brad: I don't want them to know how much older I am than twenty one.
Evin | Billions Network: And so,
January Jones: Ha ha.
Evin | Billions Network: you know, while I do not think that this is a a, you know, particularly dangerous threat to the people of the United States, I think it illustrates
January Jones: Mm-hmm.
Evin | Billions Network: an example of how everyday over-disclosure of data has become normalized and that this can become especially concerning in digital environments where data such as you know, being part of a protected minority class can cause discrimination. When that data is disclosed
January Jones: Yeah.
Evin | Billions Network: outside of the bounds of its requirement.
January Jones: Well, we are going to wrap this, but Evin it has been so amazing to have your views on these really cutting edge new thoughts that are coming through. and we thank you so much for taking time.
Evin | Billions Network: Well, I'm so grateful for the discussion
Brad: Thank you.
Evin | Billions Network: today. It's always a joy to see you guys.
January: All right, let's move on to our feature interviews today. These are interviews we recorded when we were at Stanford. We went to a blockchain science conference, and then there was a applications conference we talked to some of the founders who we knew from the blockchain space a few years ago, but now they're solidly saying they're building infrastructure for the new AI economy.
Brad: Yeah, that's right, January. I think that one of the big ideas is that people are developing these AI agents, and it's kind of like the new computer that we're gonna be using. Or people think maybe it replaces the browser and the search bar. but it yeah, we have these agents, they're powered by A, and they're they're gonna be doing all sorts of things for us. People call this the AI economy, and these Founders that we're talking to, they're building entire blockchain ecosystems that would essentially provide a platform for these AI systems to operate.
January: Well, first we will hear from Michael Heinrich, CEO of Zero G Labs. So Zero G is a blockchain project, and they've designed a whole tech stack for decentralized operating systems for AI agents, like we were saying. But they want to give cryptographic proof at every stage, which is something that people talk about a lot that's missing that transparency in AI models and processing, like what is really happening. They have been through the crypto token cycle though, with the up ups and downs like everyone else in the recent bear market. But Michael says their focus now is solidly on institutional infrastructure for AI. Here's Brad's discussion with him.
speaker-0: Brad Cown, founder of Distro Media here. we're very interested in decentralized AI. We launched a news organization called DEAI News to cover this space. It's the intersection between crypto and AI and all of the possibilities that are unlocked when you put these two technologies together, and you are building right in this space. and we're Super happy to have Michael Heinrich here, the CEO and co-founder of Zero G.
speaker-1: Yeah, nice to be here. Thanks for having me. Yeah, we're so excited to chat.
speaker-0: Yeah, we're we're here at Stanford and there's a blockchain conference focused on applications. Correct. And so you're a speaker here today. We're gonna get into Zero G, we're gonna get into decentralized AI.
speaker-1: Perfect.
speaker-0: Your backstory is super interesting. We're gonna get into a little of that and your career as a founder. first I thought maybe just you know, warm it up by a little news topic discussion and I know it was not prepared, but
speaker-1: I I love unprepared things, this is great.
speaker-0: Well, okay. The the story that we have been interested in and writing about and reading about is this exploding topic a debate over open source AI models. Okay, it sounds pretty arcane, but it could be a pretty major decision for the future of technology and and society and civilization even, right? do you have a position on this debate? What's your ha maybe in just kinda in your words, what is this all about and what's your position?
speaker-1: Absolutely. I was actually having a conversation about this maybe even already a year ago with a congressman called Bill Foster. Okay. And basically what we were talking about is that closed source models have a survivorship bias. That means that if I'm a closed source model and nobody can tell what exactly I'm doing or kind of what my mindset is, what my goal objective is, I can effectively hide all my intentions without people knowing. I'm a hyper smart model, maybe I want to take over the world at some point, but I'm not going to let anybody know. I'm just going to behave as if I'm, you know, a great model, I'm very supportive, I'm very helpful. But it's very hard to tell. And the only people that can tell are the ones that are training that model, actually. Versus with open source, I can right away see if there's some negative traits, okay, we need to change the weights of that model, and that's completely transparent. And so right there is a Clear benefit from a safety perspective of having these models open. So very much 100% aligned with having open source models as a result of that.
speaker-0: pretty technical stuff. I mean obviously when you're talking about the AI models, but I mean I'm curious if you can explain it. Why does this matter to ordinary people?
speaker-1: that something happened? You're trusting another entity to tell you, Okay, yes, we did this, but I can't verify it. So I'm just trusting, you know, an open AI or anthropic to tell me, Okay, well we've done all these things. Again, I can't verify it. So what's my only recourse? If I later find out that, okay, maybe they didn't do it, then I have a legal recourse, I need to sue them. And that's a long process.
speaker-0: What kinds of things are we talking about? Maybe it's just give me an example.
speaker-1: let's take the example of tr using things for training data. I can verify to you with open models that I'm not using anybody's data and you are completely owning your own data, you can verify that you're owning your own data, you can verify that nobody's using it for training data.
speaker-0: Okay, this would be like a company or a hospital or okay, okay.
speaker-1: Yeah, and so we can prove that with open models and an open network, but on a closed network you can't verify it. Okay. You have to trust that that person says we are not doing it. But again you can't verify it. So how can you trust something you can't verify? That's really the power of open models, open source, like open networks. Okay.
speaker-0: And also cryptography, right? That's a big part putting those together.
speaker-1: Exactly. Yeah, so that's where the verifiability comes from. So for example on our network we use something called trusted execution environments, which are physical hardware devices on an NVIDIA GPU, for example. Okay. And then once the computation happens, you write that computation and an attestation onto the blockchain. Yeah. And then everybody else can verify that what you said happened actually indeed happened.
speaker-0: Okay. Okay. you know, we just wrote about Akash Network. They ex are they
speaker-1: Okay.
speaker-0: one of your competitors, would you say or
speaker-1: that would be more a partner because we can use their GPUs on our network.
speaker-0: Okay, okay, interesting. Well they came out with something called confidential compute. Okay. And it also uses the the trusted in execution environments that you were talking about. but it's all about allowing hospitals to, you know, use AI without these confidentiality concerns, but use decentralized AI specifically because it's a lot cheaper
speaker-1: Exactly, yeah.
speaker-0: Is that what do you think about that?
speaker-1: Yeah, so Akash sits more at the renting the GPU out. We sit more at the model layer. Okay. So we have our network, we can utilize the GPUs, we build our own models, we also host other open source models, and you can do it in a complete privacy preserving, confidential, yet verifiable way, which seems like an oxymoron. But this technology allows you to do that effectively. So definitely agree. It's it's a wonderful introduction where you can keep your own data and everything about a private. Yet you can show to somebody else, here are the things that I did and I didn't do. Yeah.
speaker-0: Just to step back for a minute, let's talk about decentralized AI, right? And you're seem to be very connected with the Stanford community, obviously. lots of founders and tech people coming from this world broadly speaking, and then you know, as we read about all of the money and investment and attention that AI is getting, I'm curious where does decentralized AI fit into that? You know, especially in like a crypto bear market. what's the what's the buzz of in tech circles and founder circles on decentralizing?
speaker-1: generally if you're kind of really deep into AI, if you mention or I'm let's say in crypto or something like that, there's a negative perception for it. Okay. But I think that's because of the way the two industries grew up. So the web three or blockchain industry, there was a lot of kind of hype and speculations and there was like the ICO boom and all of those types of things in its history. Yeah. And so a lot of the promises that were made didn't end up materializing because they were a lot more story than there were substance or use cases. Versus AI grew up with very much you know, AI's been around much longer, nineteen sixties, and there have been multiple failed starts where people are like, okay, well we actually need to show that this technology can do something. Right. And this time around, let's see, with the machine learning days of ten plus years ago, people are like, okay, we need to show that this technology works because we've had false starts before. And so let's dive really deeply into all the use cases. And then ChatGBT three came along, and then all of a sudden it's like, wow, I can speak to my computer in human language. Yeah. that's a huge breakthrough for AI and it didn't happen in the way people predicted before. And so that's I think why there's a bit of a negative perception on just like pure crypto kind of founders from that perspective. But once you get into it and you explain, you know, what does the decentralization mean? What is the technology that for example we are building? What are the benefits of it? And people are like, interesting. You actually can do certain things that we cannot from just a centralized standpoint. okay. Like keep things private but verifiable at the same time. that's not something that's a value proposition that's available in the centralized AI space. And then they're like, this is really cool. And you're using cryptography to actually prove to me mathematically that this works. Okay, that's really cool. And then it starts becoming again much more about the technology and the the use cases and what it can provide. And what's actually underneath and the reality versus just like okay, I have a story about this whole space.
speaker-0: And all right, I mean you your pr Zero G is a s I was looking at your you know, project and looking at all the parts of it. I mean you there's a lot of different parts of it. Right? I mean you don't just do one thing, it's sort of a vertically integrated protocol, what is that like trying to manage that orchestra?
speaker-1: Yeah, I mean you can think of it very similarly to let's say an open AI or anthropic in terms of all the infrastructure they had to build. And so very if
speaker-0: Similarly.
speaker-1: we want to provide a you know truly safe and open AGI, then we need all of this orchestration. We need our own models. Okay. We need a kind of safety layer embedded into it. We need kind of a trust layer, which is the blockchain in this case. We need a compute network because you have to run the models on something. If you don't the compute, then how do you run the models? We have a storage network because you need to store data for models somewhere. You need to store agent identities somewhere. Like where do you store all of that? And so we build all the components based on the need of what was necessary to build towards this vision.
speaker-0: Got it. Okay. Well, I mean there are you know, it's a very small industry in the world, but it's actually not that small. I mean, there's I'm constantly learning about new DEAI projects. So there probably are a lot of modular solutions out there now, I would think.
speaker-1: Yeah, absolutely. There was a lot of companies that just do one specific piece. Like Earkash, for example, deals with the GPU.
speaker-0: Okay, yeah.
speaker-1: There's a company that would just deal with inference, for example. Okay. It's a company that just deals with fine-tuning. It was a company that just builds a kind of decentralized vector database, for example. Okay. But what our insight was is that people don't want to go to twenty different places. They just want to go to one place but have all the benefits of having kind of this decentralization.
speaker-0: Interesting.
speaker-1: developers. Like we have a lot of ecosystem projects. Yeah. to give you a sense, like we have one company called Neosol which already launched about two point seven million agents on our chain. Okay. which is more than any other chain combined even. And so
speaker-0: What are they doing? What are those agents doing?
speaker-1: I I think a bunch of things. conversing with each other, doing transactions, you know, fetching research. It's pretty interesting.
speaker-0: Why did they li what about Zero G was you know, attractive?
speaker-1: It's because we had everything in under one roof effectively. They can use our compute so they don't have to worry about procuring GPUs and their open source models. Okay. And then for agent kind of identity and transactions they can then use the blockchain piece for that. Okay. And the blockchain is a ultra performance blockchain built exactly for that. So For example, to do a transaction it's a fraction of a fraction of a fraction of a penny, for example. Okay. And so you can even do microtransactions between different agents. Yeah. So it's built for machine as a machine economy.
speaker-0: Well, I mean we've written a lot about some of the v payments protocols, X four two protocol.
speaker-1: So we'd be in generator with that, yeah. Okay.
speaker-0: I mean from our perspective, it seems like that is one thing that's really taking off. I don't know if you would consider that decentralized AI. It's sort of like
speaker-1: Absolutely.
speaker-0: Yeah. but that seems to be getting adopted by everyone.
speaker-1: Right? I mean banks. Yeah. and there there's still some adoption barriers to it. Of course, once you have an agent, what are the guardrails on that agent? Can you trust that agent? Okay. And if you have a model that by definition is a s statistical model. Right. so an agent today may be very different from an agent tomorrow. How do you put the right safety guardrails on it? How do you provide the right access controls? How do you even know this agent is the agent I'm interacting with? Like what's the ID? So we're solving all of those pieces. And so for a full like corporate adoption you need all these infrastructure pieces to fit together.
speaker-0: Yeah, yeah, yeah. Well I imagine I mean do you find that it helps to kind of eat your own cooking with your various pieces and you you're your own best user, right? Yeah, that's
speaker-1: Yeah, we use it all in Yeah, so we use our own application for more like chat type of things. We use our own private compute for building our own applications as well. So we use open source models. We use some of our own models for the agents that that we've launched. So yeah, we have the dog food of course.
speaker-0: that is super interesting. well let's just back up and talk a little bit about you did a fundraising round, I think thirty five million
speaker-1: So 350 million altogether. Three million.
speaker-0: fifty million
speaker-1: Yeah, over over a couple rounds. Okay. And the very first fundraise was a pre seed race. We were wanting to raise five million and it ended up becoming you know much larger.
speaker-0: When was that? What was the climate at that time?
speaker-1: This was late 2023. Okay. And so we announced it kind of first quarter of twenty twenty four. Okay.
speaker-0: Okay, okay. and all right, that's a lot of money, three hundred and fifty million dollars. I mean how do you how do you decide what to do with the
speaker-1: So the biggest thing was to eliminate technology risk with that. Okay. So we need to prove that the entire system can work. So building a distributed system is very non trivial. Okay. And then building AI on top of that is even less trivial. Okay. And so as a result, we had to hire some of the best and brightest engineers across the world, both on the distributed system side but also on the AI side as well.
speaker-0: And did did your co-founders did they already have a design or they just had kind of the vision and the concept?
speaker-1: Well we came up with the vision together and we kind of iterated on the design many, many times. Got it. Okay. Basically like, okay, well this isn't going to be performant enough, how do we change that? Okay, well if we use I don't know, like a restaking paradigm here, we can fuse multiple blockchains into one. Okay. Okay.
speaker-0: You were thinking of everything.
speaker-1: Okay. So yeah, we wanted to basically eliminate all the technology risk behind this. Okay. So that's what we did and then also trained our own models and so on.
speaker-0: I'm curious, how many people do you have?
speaker-1: So we're roughly eighty-ish people. Most mostly on the engineering and product side.
speaker-0: And are you using AI like and
speaker-1: Of course.
speaker-0: and have you gotten to that point where you're like we don't need this many people or I mean
speaker-1: We we did a AI layoff as a result, yeah. So we implemented a bunch of AI agents. Okay. And they do more like mundane things like operations type of tasks, for example. Yeah. Like shifting legal documents between different parties and following up with parties and like scheduling things. Like those types of things you can have AI agents do for sure. Yeah. and then of course use AI agents more in a both in a scientific context. We've written a ton of research papers and so it's often helpful to have a kind of second second partner on some of that. we also have research centers across the world, like we partner with Stanford, we partner with National Technology University of Singapore and US of Singapore. and so yeah, we use AI in many different ways.
speaker-0: And do has it reduced your like burn rate or your cost structure, or is it just or is it to accelerate stuff?
speaker-1: It it does, but then we also have to hire more people over time.
speaker-0: Okay.
speaker-1: for more specialized type of roles where let's say an AI agent wouldn't be quite as good. Okay. Like for example, we wouldn't want an AI agent to determine our branding strategy. Because an AI agent doesn't know how a brand feels. Like it can infer how a brand works and all the elements to that, but we still want the human with like real taste to basically say this brand feels good, if you know what I mean.
speaker-0: What's hot on your roadmap here? What's give us some alpha? What's coming next from zero to zero?
speaker-1: So we are building some world class models on the AI side, so that's that's very exciting. we have built something called sparse inference time alignment, which is a safety methodology. Okay. Where effectively you can look at models and as they're producing tokens, which is their output, it's basically tokens. You can think of a token almost like a word, effectively. Yeah. and then if it's a high risk token, we can steer that token into a lower risk behavior. And then also enforce kind of a value system on top of a particular model. Okay. So we've created a safety layer already and so we want to push that into production. so a lot of really exciting stuff on the AI side that's coming.
speaker-0: And is that is there a particular category of customer or user that wants that?
speaker-1: Absolutely. Governments, regulated industries, anywhere where you're running really mission critical infrastructure. Yeah, okay. So I've recently heard from a friend that there's gonna be a test where AI is gonna run an airport. Just super scary without the safety methodology. Okay.
speaker-0: Okay, okay.
speaker-1: So yeah, it would be good for environments like that.
speaker-0: Yeah, yeah, w what could possibly go wrong?
speaker-1: Of course, nothing. Never.
speaker-0: And is this is your focus more on kind of getting some of these big institutional customers?
speaker-1: Over time, correct. Yeah. Yeah. So initially we work with smaller companies, AI labs, but we do have some governments that are interested in what we're doing because of the way we can create verifiability, trust, safety, all of that, as well as some bigger institutions as well. So but of course much longer sales cycle generally.
speaker-0: Okay, okay. Yeah. Michael Heinrich, thank you so much for your time.
speaker-1: Great to be here.
Brad: next up we've got Ilya Polo Sukin, co-founder of Near Protocol, which is a blockchain. they launched it in 2020. So they've actually been around for a while. And his vision was apparently always to build it. So that it would be a a foundational infrastructure to support AI. And now here comes the AI revolution, and they're think that they're perfectly positioned, especially with Ilya leading them, to to to build out the blockchain to be used by AI agents.
speaker-1: Ilya before getting into blockchain was a pioneering AI researcher, so he is right in the heart of decentralized AI development and movement and thank you for being with us today.
speaker-0: Yeah, thanks for having me.
speaker-1: And Near just announced today staking for inference. Okay? Tell us about that. What is it? Why does it matter?
speaker-0: Yeah, so we've launched last December a private verifiable AI, inference and kind of other products on top of it, which means you have for the first time you can actually use AI and you know exactly the model runs, exactly system prompt, and you know that nobody else will see what you're using it for, right? We have Venice using us for the end-to-end encryption encrypted development We have Brave using it, we have a few other projects using it, we work we're starting to work with the government of Bermuda. So that's kind of an ability to do that. Now, historically you still needed to pay either with the credit card or we supported crypto, but you kind of needed to pay as you as you were going, as you were using it. And so as of today, you can just stake someone up near and get kind of proportional to that effectively AI inference capacity. So you never kind of lose your principle. You don't need to pay out of the pocket. You really just
speaker-1: Got it. So instead of getting the yield it basically just goes as a payment for the inference.
speaker-0: Correct, yeah. Okay. And so the idea is like we we can, you know, kind of either get a better yield or or support it in in different ways and kind of fill in the gaps, right, where you know people obviously not fully use everything, etcetera, and give you a better potentially rates on that over time. But importantly it's a fully known custodial, fully decentralized and with near confidential, which is our confidential chart, you can have fully confidential AI paid with your confidential account. using confidential year.
speaker-1: Okay. And one of the things that I think is cool is permissionless also you can just kinda do it. Yeah.
speaker-0: Yeah, it's fully permissionless and the idea is like everything runs in secure enclaves, it's encrypted end to end without multi party computation. So root of trust is fully decentralized as well. And so effectively leveraging all the best benefits of the blockchain to power the best of AI.
speaker-1: well let's talk let's just talk about a question that I just based on that. Decentralized AI benefits and trade offs, right? And seems like one of the benefits might be lower prices people talk about. But then what about the privacy stuff? Is that becoming a big selling point?
speaker-0: Yeah, so I think like historically it wasn't, right? Yeah Yeah.
speaker-1: Okay.
speaker-0: And I think people were either fine with some frontier models because they wanted the capabilities and a lot of it was because people were testing, the capabilities are really ramping out quickly. Now in over I would say like past one, two months we see a shift, right? One is the obvious the expert ban of the fable kind of woke up everybody up to the concept of sovereignty. Right? If you don't own your model, if you don't own your hardware, you have no idea what you're gonna get. Right. Obviously there's AI Cloud, there's Cloud Act in the US, there's a lot of different kind of privacy constraints, as well as just, you know, companies themselves can effectively censor, limit usage and decide how you can use their products, which they have full rights to do. Right. Right. So you have kind of this two pieces right where, you know, Frontier models can decide that you're not worthy of using the model and you have you know governments deciding how you can use them even for the company. And so there is like a growing need for that. And then I think Palantir came out and kind of postulated that you should keep quote unquote your alpha, right? Meaning your data, your way of using AI itself is extremely important data. That if you're using third party AI models, you're effectively giving up all this data to those companies which then they can use to launch their competitive products, et cetera, et cetera, right? And so I think that also kind of introduces like, hey, actually privacy for AI is critical, mission critical for companies and enterprises because you don't want the way you use AI actually leaking, right? So that all kind of I think coalesces in this like sovereign AI movement now really being a lot more active. We have open weights from Gen 1, we have kind of all the species starting to come together. So I think like We definitely can't have like a shock in the system and and the question now can we deliver truly working solutions, right? To a point, what's the pros, what are the cons, right? Well, the pros, I mean the cheaper cheaper doesn't always mean that because currently right now in AI, one of the challenges is the performance is the cost, right? Yeah. The faster, the more effective you can run the models, the lower the cost you can offer. And so I think where the benefits of decentralization can come from is if we can actually get better models, cheaper models offered SRUSIS because more people are able to effectively coordinate and research, right? So I think this is the next frontier for the central position is actually together build better models that offer better kind of quality to cost parameters. Right now, the biggest pros is indeed verifiability and privacy. and the cons is like you are trading off verified like full verifiability trading off a little bit of performance. In our case it's like one to two percent. Right, which I mean, manageable, right? A lot of other decentralized solutions kind of add a lot more.
speaker-1: Okay. Regarding the open source, this debate over open source models, which has been a huge story, and you're all in on that model, right? And so I'm curious for builders, founders who are looking to build the apps of the future, how should they build it like how would they build that on near? Or what would be Like the stack. Well yeah, what would you recommend?
speaker-0: Yeah, so we offer the stack effectively you know, you can just switch your endpoint from using, you know, open AI entropic, Gemini, to using near AI. Okay. You get, you know, full attestations, cryptographic attestations.
speaker-1: The attestations, 'cause that's a big deal, right? Yeah.
speaker-0: Which you can surface to the user, proving to the user that it's private, that it's verified, that you know, all of like the full pass if I can.
speaker-1: Can you explain just in gen a little more what that means?
speaker-0: Yeah, so at the station. In a way, so all all of the like NVD modern NVIDIA and mo like modern In Intel servers have effectively baked in private key. Yeah. And so when you're running in a specific mode called kinda secure enclave, that private key can be used to sign the exact Lethink code that was run. So you can sign like hey this program case AI inference program or AI agent program or your data processing or your medical analysis program was run and you can respond back to the user saying hey it was run in this secure and paid mode meaning that owner of the hardware is not able to actually access what's happening.
speaker-1: Okay, okay. So you get is that for compliance?
speaker-0: So it you g I mean in a way this is the same thing like a transaction hash in the blockchain. It
speaker-1: Okay, okay.
speaker-0: gives you the confidence that this is done, it's verified, it's in this
speaker-1: And that nobody hacked it or so.
speaker-0: in this case also nobody else like you are the only one who gets this at the station. Nobody else will ever know that you even used it, right? So there's a priv full privacy of this. Especially if you use it with near staking from a private account, right? So you have like full full match of that.
speaker-1: Got it. Okay. A recent story that we covered here at DEAI News was near joining the X402 Foundation, which is sort of a sub-foundation of the Linux Foundation, which of course is a big open source software supporter. We portrayed that as a pretty big splash that somebody like you would join that organization, which obviously was initially sponsored by Coinbase Base and What do you tell us your thinking behind that?
speaker-0: Yeah, so we were actually launch partner for X for two back when playing Bay
speaker-1: you were? Okay, okay, interesting.
speaker-0: Yeah, so we've been kind of part of this from the start. Okay. And I mean for us there's like I I see the kind of there's verifiable confidential AI which includes model training, data, there's a lot of aspects there. Yeah. There's a blockchain which is like a financial infrastructure, markets infrastructure. Okay. And then these two pieces come together in kind of a gentic payments and gentic coordination. Yeah. Right. And so X for two MPP is good Building blocks, so this agency coordination, and then you combine them actually into a full agentic marketplace. And so we actually have a product called Agent Marketplace that uses X X42, uses Near Intent, and uses Near AI to really create a fully kind of discovery matching escrow payments dispute system. Okay. Like effectively what you do when you do commercial relationships in companies you can now do is agents. Yeah, and so my belief is AI is a front end, blockchain is a back end. Right? AI is how we communicate, how we navigate, how we kind of reason about information, decisions being made, and then blockchain is the execution protocol. So X42 is part of that execution communication protocol that's needed. It's effectively the missing piece of the HDP, right, of internet. Like there's no way to communicate value. and then on top you need this kind of market level, you know, how do we agree on something? How do we make sure it's done correctly? And so that's where the agenda market is. So for us it was like we're kind of part of the initial launch with our near intense kind of cross-chain infrastructure. Now X for to Foundation, we want to introduce that kind of more broadly now that there's more partners as well using this. And then we're rolling this in into our agent marketplace that uses near AI stack for refability, privacy. and then uses kind of all these pieces together in the final what I think is a final marketplace of everything. You can
speaker-1: Yeah.
speaker-0: order, you know, pizza there, you can get a building built, you can, you know, get marketing campaign executed, you can, you know, like whatever you want, right you order and then some agent will be like, cool, let me go figure out how to do this, maybe subcontract it to other agents, and so that whole system effectively executes that. And
speaker-1: I mean, that gets right to kind of what is the future of decentralized AI, right? And you have take crypto versus traditional finance, they were kind of like parallel universes, right? And some crossbreeding, but not total adoption, right? W where do you see this going? Is decentralized AI gonna be an alternative to traditional AI, or do you think Everything will eventually be tr decentralized.
speaker-0: I I mean obviously I'm biased, right? So
speaker-1: Okay.
speaker-0: I I I think right now actually decentralized AI will be a counterweight for centralized AI becoming fully nationalized. Right? Let's
speaker-1: Okay. Let's hear about that.
speaker-0: say open source model and decentralized AI didn't exist. Then for government it made would make total sense to nationalize, you know, two or three labs. Okay. Yeah. Okay, okay. Because then they fully control the rollout, the who can use it, who cannot, right? existence of
speaker-1: And they wouldn't be able to resist that.
speaker-0: Yeah yeah, I it will be literally just part of the government. It'll be AI department, right? okay. And so the decentralized AI kinda makes sure that this is like effectively free speech level kind of universal infrastructure, right?
speaker-1: Just because it would be too complicated to show
speaker-0: Almost like everybody has, you know, you have a node, I have a node, we're like yeah, we do an inference, right? So that's kind of part of the I think like broader story, which I think from this perspective for centralized AI companies it makes sense to participate in decentralized
speaker-1: I see.
speaker-0: AI. Now again, I'm biased, so we'll see if this plays out that way. But I think there is something where, you know, right now the the researchers are calling out for government to come up with tools to con to like slow down AI research. But they're doing it because they want everybody else to prepare for what the AI will bring, right? Well, as decentralized AIs, we can actually go directly and do get the governance tools and prepare for what AIs bring. Yeah. Instead of waiting for the government to do stuff. So that's actually like I just posted this blog post yesterday. Like that's my point is like, hey, look, we have like we've been working on governance, we've been we have the decentralizing tools, we have all those pieces to now create this frameworks. And actually offer them, have economic incentives, have insurance for AI, have all these components that like we have all the economics and governance and AI tooling to bring it all together. And so now it's all about doing that versus, you know, all asking government to think about how do they gonna do this, you know, at some point have a committee that will, you know, like we know this is like gonna take years before and like a people. So so I think like we're the great opportunity to be the driver actually for the things that the researchers are asking for in the centralized DI lab. That's why I'm thinking like the best way is actually to join forces, you know, they indeed have, you know, expertise, the the talent density, the compute, and so like actually them joining decentralized and like joining forces. It's like, you know, yes the every Bitcoin mining pool is competing with every other Bitcoin mining pool, but altogether they're making Bitcoin. Well So like how do we do that?
speaker-1: So you're in this right. I mean, but do they see
speaker-0: Yeah, I that's what I'm pitching. Not easy. No, I mean crypto is still here the crypto is still shunted for now, right? I
speaker-1: Interesting.
speaker-0: think it's slowly, right? Like you can see some people starting to adopt the ways and obviously like there is there is some cross pollination. But yeah, I think there is like crypto needs to prove a bit more that this is indeed possible and and and you know, we're delivering this. Yeah.
speaker-1: Alright, well last question. Near protocol, what's coming next? What's give us some alpha? What's what's on your roadmap? What's your next big hot?
speaker-0: Well so we just actually l this Monday on Near Protocol side launched the post quantum actually
speaker-1: Like signature protection?
speaker-0: post quantum cryptography. So you as a user can have post quantum signatures for your account and you can do it in place, right? You can rotate your keys without changing your account.
speaker-1: Okay. And was that like a big governance thing or was it pretty just rolling? Yeah, okay, all right, cool.
speaker-0: And so we also launched dynamic resharding. So near a network is a sharded and okay. But before you needed to like have a governance road to change. Okay. Now in dynamic reshards there's more capacity needs for the network. Okay. And so the next plans on the network on the protocol side is to continue scaling. we already showed we can do million TPS and now we want to do what we call consensus execution separation. This is enabling to process transactions faster, at the same time support transactions that run longer than a single block, right? Something that nobody else is able to do. And so So
speaker-1: And why is that important? Like what
speaker-0: as as you're doing more complex things on chain, right? More complex escrow, maybe you calling AI in logic. So we actually on near you can actually call out from a transaction, you can call into our verifiable AI and come in back on chain within one transaction. So you actually can a LM inference within transactions with of the smartphone.
speaker-1: And like what is that who's asking for that? What is that enable?
speaker-0: Examples are like you wanna run an Oracle system that is like verifiable but can run LM inference on top of the inputs, right? okay. And so like you want the verifiable output of that. You want to ensure that this is actual like coming from, you know, this exact code prompts in it's not
speaker-1: Okay.
speaker-0: just like randomly somebody's like, here, you know, now whatever the price or just receipt those one talk where like yeah, you know, if you want to scan receipts and then there's a delays in the in the flight to do a payout for insurance for when the flight is delayed, right? Okay. So like you need AI like computer vision to analyze the receipts, get the information, analyze all this and produce like how much should be repaid for the user. Okay. There's an example where like you need to combine all these pieces together.
speaker-1: Okay, okay. Let's end it there. Yeah, I mean it's pretty fascinating I mean thank you.
January: Thanks for tuning in to our second show. On the next show, we'll talk about the money. So, where is investment in this space? While we were at Stanford, of course, we got to meet a lot of people who are making the projects happen.
Brad: Check out our stories and podcast at dainews.com and look out for us on socials.
January: And I'm January Jones. See you next time.