Scott Stuart - Kava Launching DeAI Attributes Beating Big AI - TOKEN2049 Dubai 2025

What I'm going to talk about today is how decentralized AI is changing the power map of how models are delivered to users around the world , how we're transitioning from large-scale enterprises to a growing open AI ecosystem, and how decentralization can help with this . The main thing that Kava will announce at the Token 2049 event in Dubai is that we have created a new upgrade for our current Kava AI model . It outperforms OpenAI, Anthropic, and Deep Seek on specific on-chain benchmarks , which we will look at in detail. Plus, with this privacy feature , it has already crossed 1,00,000 users . Let's take a look at where those users come from and how it's growing .

First, let's take a look at an overview of what kava is. Kava is project launched in partnership with Binance in the fourth quarter of 2019. It comes from the intersection of the It comes from the intersection of the Cosmos and Ethereum environments . Many of the AI ​​tasks currently being undertaken by the Kava ecosystem build on the success it has already achieved in the DeFi sector . Kava has formed some of the largest alliances in the DeFi industry .

It includes native USDT integration. Every USDT you see on Cosmos chains like Atom, Sei, or Injective was actually created on the Kava EVM and transferred to that chain . Likewise, we have an exclusive partnership with Tether to do this , and we have also partnered with BitGo for WBTC. So, all the native WBTC you see on Cosmos chains is issued on Kava and converted via IBC . Those assets and many others are integrated with the world's largest exchanges like Binance, Bybit, Coinbase, KuCoin .

This enables users to transfer these assets on the Kava blockchain between various exchanges and third-party custodians such as Fireblocks and Bitgo . As I already mentioned , Kava Diffie has been around for a long time . This is the only exclusive partnership with Tether within the Cosmos ecosystem. It has a large number of apps in use in use , and a large community has been built over the last 3 years has been built over the last 3 years . There are very few chains with such large native assets .

We have been working with Binance since the time it was called the BNB chain , so over 2.5 billion assets have been securely transferred between Kava and other chains transferred between Kava and other chains . The abundance of assets on the chain, the low transaction fees, and this native cross-chain architecture, which has been used for assets , and we can leverage it for data as we move into the AI space . As a result, Kava has multiple integrations and alliances with all the major companies in the industry , and we are now building on top of that with AI. We will provide a brief or superficial introduction to how traditional AI works and how that work is being moved to open source and decentralized spaces.

At a glance , AI is essentially comprised of an application layer, a cloud or on-prem layer that serves those models, and hardware. At the layer, ChatGPT, Claude, and DeepMind are popular applications that are powered by companies that train large language models and provide them to users with different performance . What is interesting to note is that the number of parameters for these models has increased dramatically . By about 2021, 2022 , there were enough parameters for these things to be interesting and useful , and now you can see the activities of users, and there is actually a lot of engagement and addiction to ' actually a lot of engagement and addiction to ' intelligence as a service' . The interesting thing to note is that there doesn't seem to be any upper limit to how many parameters these models can have, as long as computing power exists.

These big companies are competing to build better versions of these models that are evaluated for various things like human language, mathematics, and the feature we are releasing is Web3 or blockchain- related functionality . So, at a glance , it's clear that AI is here , and it's being incorporated into all major industries, including blockchains . We call decentralized AI or DAI, which is taking open source and open weight models and implementing them on a blockchain, and there is a huge growth in that field . At a glance , the stack looks mostly the same. You have AI dApps and an agent layer .

You have an open model layer, and these ' and these ' deep in' networks or dClouds, whatever you call them, are implemented by a full node, similar to a smart contract running on Ethereum, and provide these in a permissionless manner. DAI is a confluence of two technological ideas. When it comes to blockchains and AI models, you can't have a truly decentralized system unless it runs on an open and permissionless blockchain . We are now witnessing the expansion of the AI ​​sector . Therefore, many projects are proliferating on AI Therefore, many projects are proliferating on AI DApp and agent platforms .

Different open weight and open model projects are emerging emerging , and we are also seeing a proliferation of compute and DePIN projects that host these models . Now let's look at Kava AI . Our primary goal is to build a decentralized AI model that can provide and scale intelligence as a service at low cost, as part of the common Kava ecosystem; This is a very difficult problem. We are continuously working on this . We first launched this at the Token 2049 event in Singapore .

As a result of continuous hard work, Kava AI has now crossed 1,00,000 users. Let's take a look at where those users come from and what our value is. A large portion of those users come from all over the world, especially from Southeast Asia and the Asia-Pacific market , which we will talk about in detail. Okay, let's get to the point. What we are trying to do or working towards at a high level is to provide a utility to users .

You can go to chat.kava.ai right now, open the app , and start a conversation using a general intelligence model. You can talk to it just like you would talk to Chat GPT or Deep Seek. The main feature in it is that you can use Anonymous mode .

So, all data, information, and any weird questions you want to ask will be stored locally on your browser or phone , and you can delete it at any time. They are stored only on your device . Users can register if they wish and have that data stored in a SQL database . The main value of this is that users can anonymously interact with the model at low cost, or select the best model and store the data in an off-chain database, or pay a fee to get a more advanced model. Another important feature is that users can interact with Web3.

Okay, now let's quickly move on to the listing part. So, if you are now trying to use the largest models like Deep Seek, CLAD or SAT GBT, you can try it yourself . In fact, they cannot function or think properly in a Web3 environment. Web 3 is not a huge application; Of the 100,000 users we have, only 10 or 20 percent are actually on-chain . However, this is an important application, and the fact that no one has handled it properly presents a unique opportunity .

So, what we do is we train these models and have them do things like connect to the wallet . This allows it to be connected to an Ethereum wallet such as MetaMask, Finanse Wallet, or OKEx Wallet. It can also communicate with smart contracts . So, you can call a function and interact directly with the smart contract . It now knows about ERC-20 tokens, as it can communicate with smart contracts .

You can exchange ERC-20 tokens . You can swap ERC-20 tokens . You can check your balance on Goa Chain, Ethereum Chain, and BNB Chain . We are currently working on cross-chain support for it as well . These traditional models have an understanding of DeFi.

They know what the trading strategies are, but they can't execute them because they don't have the ability to communicate with the blockchain. Once users gain basic ability to make transactions through AI , everything else becomes easier . This will help you identify emerging trends , act on those trends, and create agents to monitor and maintain your strategies 24/7 . Like I said, it's probably about 20% of the Covesting AI app usage , but it's a robust app in the crypto world and we're seeing a lot of traffic on it. Yes.

Basically, the most important thing here is what we are actually trying to fix . You may have heard of other companies that offer multiple programs or models in a somewhat decentralized manner, such as Near have heard of other companies that offer multiple programs or models in a somewhat decentralized manner, such as Near ; Even at the SAT GBT or anthropic level, the issue of issuing tokens can be solved . I'm not referring to crypto tokens . I'm talking about tokens related to models. You can provide them at a certain cost.

The problem we're trying to solve is to offer it at a lower unit price than the cost of issuing that token to a user , so that you can make a profit on it . There is a lot of hard work behind this. The team working on this is world-class in providing full nodes , which are inherently inflexible and very difficult to manipulate. Now they are moving to GPUs , offering GPUs at the hardware level , selecting the right models , training them to work with real Web3 products , offering those models to users at an economical unit price , and looking to scale it up . This remains an open question that many companies have yet to resolve, even in Web2 .

The next most interesting topic is, how do we evaluate these models ? Our primary goal was to create a benchmark to evaluate how well a model implements Web3 transactions , but we may have noticed that OpenAI recently released a version of ChatGPT-4.0 have noticed that OpenAI recently released a version of ChatGPT-4.0 . It had such huge usage that even founders like Sam didn't have a formal methodology to evaluate a new feature when it was launched .

Is it better or worse than the previous model? In that recall they did, withdrew it because there was enough feedback from users that it was bad. Anthropic also recently withdrew another version. So, beyond increasing the use of this product in the Cava Web3 ecosystem, two key areas we are focusing on are how we can issue model tokens at a low scale . Low unit cost that can be scaled on the blockchain.

So, this is a difficult problem, and we are trying to solve it . Furthermore, the next step is how to create proper evaluations to find out if this model works well for users to interact with Web3. So, I think this is a very early stage . There is a lot of work going on , but it is notable that it is receiving a good response from users . They want to use simple apps anonymously or interact with Web3 .

Although we are currently losing money and not making a profit in providing services to those users , we are continuing to work towards achieving a point where its economic operations are profitable . This will ultimately be very beneficial for the Kava chain. In short , this is a collection of the most important tasks we have been doing . Thank you for listening.