Oasis | AI without the Black Box - Matej Janež | ETHDam III - 2025

Welcome to East to Ethan. To we're going to be having I want to say it correctly. Mache mate mate mate mate talking about AI without the black box. Now yesterday morning we had an opening about AI that definitely was in a black box. So very interested uh for this and so please give it up for Mate.

Thank you, JJ. Good to know the talk's not PG-13, so I can curse like hell. You know, it's not a problem. I'm joking. I'm joking.

Um, let's just see if this works. Extended. This is mirror sometimes. Okay, we're not there yet. Just give me one more sec.

Let's see if this works. Come on. Okay, I'm cutting it out. Okay, cool. Yeah.

Hi everyone. Thank you for coming here. So, I'm Mate, the head of partnerships at Oasis. And today we're going to talk yeah we're moving slightly away from the cyberpunk privacy but still uh we will be talking about confidential compute private compute um a little bit more on the AI side which is kind of still the hot topic also in crypto and AI without the blackbox. I think it's uh quite insinuating what we're going to be discussing here.

So if you go through a quick outline we're going to look at the current AI landscape. We're going to speak a little bit about the differences between centralized decentralized AI. Then we're going to move to the Oasis raffle framework and we're going to finish it off with uh WT3 and I'll talk more about what it is uh towards the end of the presentation. So current AI landscape uh early AI right it was already present in the '90s. Uh it was called more machine learning.

Now machine learning is kind of more of a subset of a broader AI uh field, right? But we are reaching now that tipping point where we can see AI being everywhere, right? So we have everybody using LMS for different tasks. We see now agents coming in, you know, uh taking over tax tasks for people, people saying, you know, that they that the tech companies are laying off people because they can do so much with AI. So uh I think we are at that point where we can see broader adoption of just general AI tech.

uh but we can also see that we have the power concentration coming to into the hands of the few right we have the acceleration of that compute power and capabilities coming to Google entropic open AI meta uh and because of their capabilities everybody else basically get getting pushed out away but there are alternatives right we we are building the decentralized options they're in development but I think there's still a little bit of a lack of understanding from the user perspective why use decentralized options, right? If that just works, why why go the decentralized route? And if we look at uh we if we can just do a quick comparison between the two, right? Centralized AI side, as I said, dominate dominated by the large tech companies and you have the data compute power, development talent in the hands of the few. Of course, there are advantages to that model, right?

You have economies of scale, right? They have access to data. They have rapid development. But at the same time you have the disadvantages which are basically all all to the detriment of the user right you have data privacy concerns you have monopolies over AI and then that enhances the risks of bias of um how how are those LMS what are those LM feeding you right um and the tech companies they they don't really care usually right they prefer to look at the new AI features than care about the data privacy of the users right uh but we do have the decentralized AI alternatives, right? Which is removing the single point of failure by adding transparency, verifiability, and users are more in control of their data, right?

Obvious advantages, control data, enhance data privacy, and of course less potential for bias, but there's always a trade-off, right, with slightly potentially slightly lower performance. There's slower coordination between all the decentralized projects building you know and the lack of standards also doesn't make it easier to cross integrate but we still we need to keep that focus on the user right uh personal control over consent indiv individual choices you know that they have their data ownership the transparency so they have clear visibility into their data use and the uh decision-m and of course in the end you know there's also value recognition um because personal data is valuable, right? And currently users are not getting any cut in in that uh in that economy, right? So, uh including the users in the data economy is definitely something that we'd like to see uh moving forward, right? And the decentralized AI landscape isn't small at all.

This uh this kind of landscape overview was done end of last year and I think right now it's probably triple or quadruple this, right? But it still holds true to the main players in the space. You have different fields right from confidential computing, development platforms, decentralized compute, agents or a very big topic. You have training, AI works, algorithms, gaming like everywhere. AI is everywhere now.

And if we look at what is doing here, right, we started off back in 2018 as a layer 1 blockchain and we had this unique layered architecture with native layer twos. We went mainet back in 2020 and smart privacy was always at the core you know from the beginning in our minds. We started with a more rust based private runtime and then we switched to Sapphire which is the EVN compatible confidential layer uh to better serve just the developers and the whole crypto ecosystem and right now we're focusing on raw raffle uh it's a runtime offchain logic framework or rolling on the full laughing whichever you it's easier for you to remember uh but yeah it's the latest tech that we're working on it leverages these for security paired with saf fire for onchain verifiability. It basically allows you know off-chain components to seamlessly communicate with the onchain rail. So you have full composibility through through the whole stack and you have flexible computation, right?

Because we know blockchains are quite limiting in in in in their compute, right? But this enables non-eterministic applications um that blockchains can handle. How raffle works, you know, we have a very nice table here that our engineers put together. I'm not going to go through this because what raffle does is obuscates away all of this right so you don't have to think about this um and if you look at a simplified view this is what you kind of need to know right and you have very simple commands oasis raffle build and oasis raw build verify to to do all of this right um so it enables basically offchain compute to inter interact with external services plus the onchain logic and if we look kind of the three main pillars, right? It does TE provisioning.

The Oasis nodes provide the intellects for raw applications. It does the onchain registrations of raffle apps, right? And the smart contract integrations. So next steps for raw um currently it's on mainet with support um supporting TDX. Intel TDX is a type of a TE, right?

Um, but right now developers still need to provision their own trust execution environments. Um, and we're building a rough marketplace or maybe a more commercial term would be Oasis D cloud where it's going to be compute providers coming into the marketplace and providing compute services. So developers only need to create like a docker compose file uh put it on the market or select a compute provider and then they have their ruffle app running right. Um next is also going to be ruffle app templates for like some standard use cases. We're going to create a template so it's even easier for the developers or even users to deploy them.

Uh we're going to see that more on the agent side of things. uh how to make it easier for people to deploy their own personal agents so they don't need to rely on third party stuff and GPU support is also coming so you're going to be able to connect GPU compute for more intensive stuff uh paired with TDX and speaking of agents right we do believe that there's there's going to be a rise of the agents and there's a lot of industry momentum here we saw I think it was mentioned by JJ that yeah he saw a lot of scam semi- agents uh on Salana towards the end of last year and yeah there was a big craze um around the the the agents but those were probably more like yeah more the chat chat bots on Twitter right um and yeah devs had full control of the wallets so they could yeah basically rug anytime they want um but even though there was a lot of hype there's still some substance there we see agents getting traction you know in the traditional space as Jensen Huang from Nvidia, he predicted that 2025 will be the year of the agents and we see entrop entropic build the MCP uh framework right so uh we believe that yeah agents are here to stay. going to be present in web two. They're going to be present in web 3, crypto. And uh I actually believe that agents will be potentially a solution to solve the crypto UX problem because uh if you're connecting web 2 and crypto agents, right?

Users can potentially stay with the UX that they're used to, but at the same time interact natively uh on on blockchains, right? Which simplifies all the onboarding that we're trying to fight all these years already. And this brings me to WT3. So what's WT3? It's the agent you don't need to trust.

Uh this is uh our latest um our latest agent that we launched uh actually just last week in Dubai. And I'll take you a little bit back why we we started WT3. We saw that there are many uh different crypto trends in uh many different trends in crypto through the last few years. But there's one constant, right? That's DeFi.

And we still believe that DeFi is one of the pillars of crypto that's going to stay, right? You had metaverses, you had NFTTS, you had gaming crazes, but everything kind of died down, right? But DeFi is there still there to stay, right? Uh I think at its peak at 2021, we can see here uh it reached like 200 billion TVL. Right now it's around 100 billion, but I think we're going to go way higher.

RWAS are not even counting here because DeFi Lama doesn't include them in the calculations. But if we include RWAS, the number will be even higher. But we can also see that AI is an emerging crypto trend. And this is also I think one that's going to stay just because of the cross um cross compatibility with the let's say more traditional ecosystem. And there was a new ter term that was coined right defi not sure how people say it right but it is uh we believe it is a winning combination for crypto right because there's the part participation in defi is still relatively low from from the users we saw how many users were participating let's say in in the memecoin craze right but so trying to hit the lottery right with gambling there but uh utilizing defi for getting more stable yields um is uh is is lower in terms of wallet diversity, right?

You have big wells utilizing it, of course, but uh let's say more retail participants are less aware of the possibilities there, right? And it could be that the users are having trouble finding opportunities, evaluating risks, right? And AI and AI agents potentially can enable uh crypto and non-crypto users uh to easily participate. What's the promise of DeFi or DFA? uh its speed and efficiency, tireless operations, 24/7 operations, complex pattern recognition due to AI integrations, nuclearization of those sophisticated strategies, but the crucial for us is the transparency and verifiability.

And this is one that we really shouldn't be forgetting about the D in DeFi, the decentralized part because with the rapid market development evolution, right? Everybody tries to develop fast and building decentralized trustless usually is complicated, right? So projects take the shortcuts to deliver and usually they take the shortcuts right uh you know where it's crucial on on the trustlessness and and decentralized part. they made a compromises there because those are the difficult parts. Uh and in the end we are still back at in the situation where users need to trust the projects won't the devs won't rug and here uh we built WT3 to kind of showcase that it is possible to build trustlessly um it's operating autonomously while proving integrity.

It's built or onchain and offchain confidential computing stacks. So sapphire raffle and it enables execution integrity data conf confidentiality and remote stations. So the components of WT3 it's OS is raffle of course is the key pillar uh key key component it uh it's for there for the integrity and fireability. It's running within the raffle uh app. We use the Imperial agent SDK which is a partner project that's deployed on Sapphire as well for the AX engagement.

Um it offers a lot more the framework itself but for now we're using it just for the engagement. We we we'll be involving the agent to do more with that framework. We're using Asian predict ocean predictor which is also a partner project that's built on Sapphire for part of the trading signals. Um and Hyperlid is the trading venue currently W3 will be trading on. Um here how it works a simple graphic right we get some uh signals from predictor uh we trade on hyperlquid we then post about the trades on on X um and then we're using the profits from the trading um to buy back and burn rows uh through our native decks Nebio sapphire um it's basically how W3 is bringing back value to the Oasis ecosystem uh through the buyback and rose buyback and uh program.

If I dive a little bit deeper into the individual components, we spoke already quite a bit about raffle. But in this case, why allows the development of custom offchain logic, right, with easy verifiable onchain um verific. And here verifiability is key, right? Because we're dealing with funds, right? Uh currently there's going to be funding from Oasis, but we're going to be opening up to the general public as well uh to participate in the strategy.

So we want to start building from the ground up in a way that it's not going to be problematic later down the line. Um the raffle framework as I said abstracts away all the complexity around te and ver verifiability and it even simplifies more with the upcoming raffle marketplace ord cloud um to make it easier for um people deploying their apps. One key feature here uh that's also um important with any agent that's dealing with wallets is the integrated key manager. So you're able to generate the wallet keys within the TE, right? So even the death never gets exposed to the private key.

I mean there's always a butt, right? And here of course if the death makes the code reveal the the the private key, it can. But as we're talking about verifiability here, the code needs to be open source. So this is the same as somebody, you know, deploying a malicious contract that everybody can see that it has an exploit there. So of course, nobody's going to use this agent if there's a there's a visible u yeah compromise of the private key.

Uh and there's an onchain log of all the onclave invocations by default. So everything's tracked on Sapphire um on predictor and the trading strategy. So it us utilizes predictor to kind of add confidence to its long and short trades. Um predictor built by ocean and deployed on sapphire and it's basically you have predictor agents um submitting predict predictions in private on sapphire and then you can subscribe to those feeds and those are the trading agents and trade off of those predictions. The current actually actually between 51 and 54% on the major pairs.

So it's not bad. um anything above 50 if you do uh quite a bit of trading is is yeah quite good and if you can manage the risk to the downside. So um have a look and try it out. Uh and then the exag mentioned uh that's the imperial agent SDK. Uh it enables us to have those autonomous AI agents with skillbased systems.

Uh it's actually integrated with race sapphire for some wraction on their end as well. Um it actually powers already the platform similarum which was quite popular end of last year. Um I think you could use similarum to create your own tokens through just twitter posts um or um yeah tweeting at similarum uh and we are looking at future integrations between Imperial and raffle. Yeah. So follow Imperial and us if you want to learn more about this.

If we look at the WT3 road map, um we're starting with the 100K initial capital from the Oasis Foundation. Uh we're going to be trading on Hyperlquid, but we're going to be expanding to other venues as we go along as well and as we deploy different strategies and as I mentioned, profits will be used to buy back and burn rows. Uh the second stage coming up next will be like a staking or a vault contract where people will be able to deposit their own funds uh and participating participate in the strategy. And then the third step is with the raw marketplace and templates people will be able to deploy their own agents uh having completely um yeah own system uh and basically deploying very seamlessly without any real coding here. So for future partners the wolf as we call WT3 is always hunting for alpha.

So uh if there are projects you know uh that offer market insights or new venues for trading or other protocols you know that uh enable some geogenerating opportunities to to reach out. Uh we're always on the hunt for that. And if I close out with just a key a few key takeaways here you know for us verifiability is key to everything crypto. uh the original ethos don't trust verify uh we hope will leave on uh the raw framework is a framework that enables you know combining onchain offchain compute in one verifiable stack and yeah we believe the future belongs to the agents they can think privately but prove their integrity W3 is an example of that showing the way so yeah thank you for that stay connected with us uh oasis.net net is the website.

W3 is the agents website or follow it on agent_w3. It's going to start trading very soon, posting very soon. And here are my telegram and x handles and my email if you want to reach out or just grab me after the talk and we can speak. And just one last thought, right, from the human torch and ironmen, the human torch is all about, you know, centralized because it's faster, right? But Iron Man is, you know, decentralized because it's better for the people.

Thank you very much. Cool. Any questions? Thank you very much. Yeah, we can put the slido up.

Um, so we're first going to get through these questions and then we're going to move some chairs at the back. Uh, so we'll do this first. So the top one or the one at the top is what are the applications of this that you're most excited about? Sure. Crypto trading is a thing, right?

I mean, is this for raffle or just general? Cuz it's for raffle. For me it's um it's going to be definitely agents because you can have so many different use cases there. Of course, raffle has other use cases beyond that. You know, you have you can have trustless oracle, trustless bridges, all that.

But to me, agents trading, defi, uh this is also, you know, where uh I got stuck in 2020, I guess, you know, never followed the other crazes, but DeFi, I think, is one of the cornerstone of crypto. So, anything related to to DeFi will, I think, be be big. Um yeah, coming up in the next few years. looking forward to it. Uh is the trading of WT3 open and verifiable?

Yeah, so um the the strategy is private. Um because of course, you know, if you put it out there, you lose all the edge, but it's containerized away from the agent, right? So you have you still have the verifiability integrity of the wallet and the agent. Um but so the verifiability of the agent doesn't change if you change the strategy, right? um because you want to ensure the the integrity of the wallet.

That's crucial here. Of course, one can say, you know, through the trading strategy, somebody the agent can lose money, right? And can argue that way. Uh yes, uh I guess so, but um it's going to be hard because people will be investing based on the track record, on the history of it, right? So, if you're going to be, you know, trying to uh have a good track record, I mean, it's it's the same with everything.

You know, there's a certain reputation you want to build. I think we saw this with Hyperlquid and their vaults and how how they're doing it there. Uh and I think it's going pretty well uh and we believe that yeah a similar system through through agents that enables multi- venue trading will will emerge here as well. All in good time. Um well then I guess how can we use the trading agent but I think you described that right?

Yeah. So the trading agent the next step will be that people will be able to deposit their own stables to participate in the strategy. Uh so that's going to be pretty on auto mode. Um but then if they're going to want to deploy their own, right, we're going to be make it super easy so they can use a template potentially modify the strategy a little bit um to their liking and then um run it within the way cloud uh as we kind of call it. Yeah.

Cool. I'm going to let you pick uh two questions. Yeah. Yeah. That's that's the lead engineer of Rafael, you know.

Uh I mean uh when we started raffle uh it was all about uh you know memes uh and it it was just a a fun one. Uh it it goes well with the runtime offchain logic but you know it's a memorable RF is kind of memorable name. Yeah. Cool. Not every like everybody likes it but yeah.

Uh any other questions on here that you want to answer? There's some down the bottom that aren't on the screen. Pump money for free. Huh. I mean the trading strategy is private so uh yeah you can deploy it but then you need to create your own strategy.

People like the t-shirt. Thank you. Our designers are great doing a great job. I know. Go off.

Okay. Cool. I'll go off in a sec. Anything else? Well, I was going to try something which was uh put up uh put up something that people can write free text in which is what are the applications of AI that you want to see based off of what you've just seen?

And if you haven't done the chairs, now's a good time to do it while we get everyone writing. And have you done it already? No. Do it quickly now. And then uh we'll see what people want to say about this and then we'll go from there.

Cool. Give people a couple of answers. But I think for now, we'll see what they come up with and hopefully that's inspiration. But for now, then we can give a round of applause. Thank you very much.