During EthCC week, at the NEAR AI Showcase, I had the chance to sit down with Illia Polosukhin, co-founder of NEAR Protocol and a significant contributor to the fields of artificial intelligence and blockchain. Known for his work on creating the transformer model and his role in adding the “T” to ChatGPT, Illia shared his vision of integrating AI with Web3 to drive innovation that benefits everyone. Illia, your work at Google and on creating the transformer model has set a foundation for modern AI. How have you seen AI change since then? It's definitely changed, right? AI went from being back-end technology that supports products to being the product itself. That's the reason I left Google originally to start NEAR AI—because I wanted to make AI the product and then leverage the data loop to improve it. At the time, in 2017 and 2018, it was too early. But today, we are at an interesting point where we have these AI-first products that are starting to emerge. The challenge with these products is that new startups have very limited time to figure out how to monetize. They raise at high valuations, they have a lot of burn, and now, many are being bought out by big companies. So what’s happening is that AI benefits incumbents because they already have distribution, the ability to spend more, and they know how to monetize user attention. Web3 is starting to solve that problem. It's giving time to build new consumer products that actually benefit users, maybe with a different business model, and leverage AI as the platform. So I think that's where it starts. There are a lot of components in the stack that will be powering that. Instead of one major company trying to do everything, you can have lots of companies working together, just like in this Web3 mindset. For a more general audience, Web2 AI startups have very little time to become hyper-profitable, or they need to get acquired by big companies. Perplexity is a great example; they spend a lot of money on compute, making their operations very expensive. That’s why they need to monetize quickly to sustain their growth, or they'll risk being acquired by one of the larger companies that can afford those costs. How does NEAR fit into all this, and what are you planning to do with both technologies? The biggest gap we saw is that there are many capable founders starting different parts of this broader stack within these big companies, but they are not well connected. There's data crowdsourcing, data labeling, decentralized inference, agent payments, and more, but there's no cohesive product. As a developer, I don't want to figure out how to use 50 different pieces—that's too complicated. Versus going to OpenAI or Google, where there's a single API to use. So, we focused on NEAR AI and the NEAR Foundation’s AI Incubator to bring these projects together and figure out an interface to coordinate them, making it simple to interact with them. Additionally, we can coordinate open-source research. Researchers at top universities don't have much access to compute capacity, but if they solve interesting problems for applications, those applications would spend significant amounts to do that. Usually, the code and models they develop are very custom-made and not reusable. We're creating a coordination hub to plug in researchers, give them credits, compute, and data acquisition, and solve specific application problems or generic problems. Their work becomes reusable by production use cases and application developers. It's like a complex, four-directional market, tying together all the disparate audiences that a centralized company would handle by hiring everyone and doing it all itself. Instead, we're creating an open platform where everything is open source and combined into one hub for everyone to use. Can you explain the concept of user-owned AI, why it matters, and how it differs from traditional AI models? The contrast here is why I don't like "open source AI" as a category. There's a lot in open source, and it's really important, but the most important part is that these models are optimized for specific functions. When I'm working at Google, my goal is to make more money for Google because that's how I'm incentivized. I get bonuses, stock options, and other benefits based on Google's profits. Even if a big company open sources a model, it's still a business decision meant to benefit that company. The opposite of that is AI which benefits every individual user. Let's say I want to consume interesting content and not get enraged all the time. For that, you need a very different model of operation and AI research. Current research happens in closed labs with defined goals. In traditional for-profit systems, there's a transition where at first you're a startup, growing your audience, and providing value. At some point, you reach your target audience or a big part of it. Now, to grow revenue, you need to monetize more from the existing user base. You've already provided them value, they're already using your stuff. Now, you need to figure out how to get them to spend more time and monetize them more. That's where the main problem happens with these companies. In tech, because of low volume costs and network effects, normally, when you hit that inflection point, there's a lot of competitors. But here in tech, you're just in a monopoly, and you start to extract value. This is where Web3 comes in—creating an economy that doesn't require you to become extractive. Crypto doesn't require posting more revenue every year. Yes, people want the number to go up, but it doesn't need to. We can be happy with Bitcoin being at, say, $65,000, and that's totally normal. We can have an economy where everybody is participating and benefiting. You don't need constant growth or expansion; you're okay with the state as it is. This is the conceptual difference we're going after with user-owned AI. What other innovations are you exploring that align with equity and sustainability? On sustainability, we've invested in being carbon neutral for NEAR, which is a Proof-of-Stake network, and we have projects for carbon credits, tracking, and reforestation. These are important. The AI researcher mindset in me is always looking at how we can develop more sophisticated AI tools to solve these problems. I have friends in cancer research and material science, and I'm excited about those fields. But where I can apply my intellect is in building better tools to help these researchers. Data scientists in these fields often don't have access to good data science or coding resources. Developers can scale their efforts significantly. It's the same with blockchain: equipping people with incentives and coordination tools to build better networks to solve problems. Do you have a favorite app or use case built on NEAR? I can’t pick favorites, haha. Every use case is great. We see a lot of user-facing products coming from Web2, like Sweatcoin, which uses blockchain under the hood for payments, loyalty, and transactions. Most of their users don't even know they're using NEAR. There are wallets becoming multi-chain, like HOT and Bitte, allowing users to transact across multiple chains seamlessly. Bitte even has a natural language interface for commands. We also have financial applications, multi-chain DEXs, lending across different assets, and more. On the AI side, crowdsourcing applications like NEAR Crowd have been running for years, improving data acquisition at lower costs. All of these pieces are coming together, and NEAR has been growing exponentially in active users. Someone tweeted that there are more side events at EthCC than blockchain apps. How can the space improve? That's why we're doing AI Developer and the AI Hub. We can try to get people to build applications or make it easy to build them. There are many misaligned incentives in the space, and just talking about them doesn't fix the issue. We've invested in bringing existing applications with users, integrating and interconnecting them, which is why we have the most users in Web3. Building consumer apps is hard, and it's easier to fundraise for infrastructure. We want to simplify the process of building applications so people can experiment more and innovate. During EthCC week, at the NEAR AI Showcase, I had the chance to sit down with Illia Polosukhin, co-founder of NEAR Protocol and a significant contributor to the fields of artificial intelligence and blockchain. Known for his work on creating the transformer model and his role in adding the “T” to ChatGPT, Illia shared his vision of integrating AI with Web3 to drive innovation that benefits everyone. During EthCC week, at the NEAR AI Showcase, I had the chance to sit down with Illia Polosukhin, co-founder of NEAR Protocol and a significant contributor to the fields of artificial intelligence and blockchain. Known for his work on creating the transformer model and his role in adding the “T” to ChatGPT, Illia shared his vision of integrating AI with Web3 to drive innovation that benefits everyone. During EthCC week, at the NEAR AI Showcase, I had the chance to sit down with Illia Polosukhin, co-founder of NEAR Protocol and a significant contributor to the fields of artificial intelligence and blockchain. Known for his work on creating the transformer model and his role in adding the “T” to ChatGPT, Illia shared his vision of integrating AI with Web3 to drive innovation that benefits everyone. Illia Polosukhin, Illia Polosukhin, NEAR NEAR Illia, your work at Google and on creating the transformer model has set a foundation for modern AI. How have you seen AI change since then? Illia, your work at Google and on creating the transformer model has set a foundation for modern AI. How have you seen AI change since then? It's definitely changed, right? AI went from being back-end technology that supports products to being the product itself. That's the reason I left Google originally to start NEAR AI—because I wanted to make AI the product and then leverage the data loop to improve it. At the time, in 2017 and 2018, it was too early. But today, we are at an interesting point where we have these AI-first products that are starting to emerge. The challenge with these products is that new startups have very limited time to figure out how to monetize. They raise at high valuations, they have a lot of burn, and now, many are being bought out by big companies. So what’s happening is that AI benefits incumbents because they already have distribution, the ability to spend more, and they know how to monetize user attention. Web3 is starting to solve that problem. It's giving time to build new consumer products that actually benefit users, maybe with a different business model, and leverage AI as the platform. So I think that's where it starts. There are a lot of components in the stack that will be powering that. Instead of one major company trying to do everything, you can have lots of companies working together, just like in this Web3 mindset. For a more general audience, Web2 AI startups have very little time to become hyper-profitable, or they need to get acquired by big companies. Perplexity is a great example; they spend a lot of money on compute, making their operations very expensive. That’s why they need to monetize quickly to sustain their growth, or they'll risk being acquired by one of the larger companies that can afford those costs. Perplexity Perplexity How does NEAR fit into all this, and what are you planning to do with both technologies? The biggest gap we saw is that there are many capable founders starting different parts of this broader stack within these big companies, but they are not well connected. There's data crowdsourcing, data labeling, decentralized inference, agent payments, and more, but there's no cohesive product. As a developer, I don't want to figure out how to use 50 different pieces—that's too complicated. Versus going to OpenAI or Google, where there's a single API to use. So, we focused on NEAR AI and the NEAR Foundation’s AI Incubator to bring these projects together and figure out an interface to coordinate them, making it simple to interact with them. Additionally, we can coordinate open-source research. Researchers at top universities don't have much access to compute capacity, but if they solve interesting problems for applications, those applications would spend significant amounts to do that. Usually, the code and models they develop are very custom-made and not reusable. We're creating a coordination hub to plug in researchers, give them credits, compute, and data acquisition, and solve specific application problems or generic problems. Their work becomes reusable by production use cases and application developers. It's like a complex, four-directional market, tying together all the disparate audiences that a centralized company would handle by hiring everyone and doing it all itself. Instead, we're creating an open platform where everything is open source and combined into one hub for everyone to use. Can you explain the concept of user-owned AI, why it matters, and how it differs from traditional AI models? The contrast here is why I don't like "open source AI" as a category. There's a lot in open source, and it's really important, but the most important part is that these models are optimized for specific functions. When I'm working at Google, my goal is to make more money for Google because that's how I'm incentivized. I get bonuses, stock options, and other benefits based on Google's profits. Even if a big company open sources a model, it's still a business decision meant to benefit that company. The opposite of that is AI which benefits every individual user. Let's say I want to consume interesting content and not get enraged all the time. For that, you need a very different model of operation and AI research. Current research happens in closed labs with defined goals. In traditional for-profit systems, there's a transition where at first you're a startup, growing your audience, and providing value. At some point, you reach your target audience or a big part of it. Now, to grow revenue, you need to monetize more from the existing user base. You've already provided them value, they're already using your stuff. Now, you need to figure out how to get them to spend more time and monetize them more. That's where the main problem happens with these companies. In tech, because of low volume costs and network effects, normally, when you hit that inflection point, there's a lot of competitors. But here in tech, you're just in a monopoly, and you start to extract value. This is where Web3 comes in—creating an economy that doesn't require you to become extractive. Crypto doesn't require posting more revenue every year. Yes, people want the number to go up, but it doesn't need to. We can be happy with Bitcoin being at, say, $65,000, and that's totally normal. We can have an economy where everybody is participating and benefiting. You don't need constant growth or expansion; you're okay with the state as it is. This is the conceptual difference we're going after with user-owned AI. What other innovations are you exploring that align with equity and sustainability? On sustainability, we've invested in being carbon neutral for NEAR, which is a Proof-of-Stake network, and we have projects for carbon credits, tracking, and reforestation. These are important. The AI researcher mindset in me is always looking at how we can develop more sophisticated AI tools to solve these problems. I have friends in cancer research and material science, and I'm excited about those fields. But where I can apply my intellect is in building better tools to help these researchers. Data scientists in these fields often don't have access to good data science or coding resources. Developers can scale their efforts significantly. It's the same with blockchain: equipping people with incentives and coordination tools to build better networks to solve problems. Do you have a favorite app or use case built on NEAR? I can’t pick favorites, haha. Every use case is great. We see a lot of user-facing products coming from Web2, like Sweatcoin, which uses blockchain under the hood for payments, loyalty, and transactions. Most of their users don't even know they're using NEAR. There are wallets becoming multi-chain, like HOT and Bitte, allowing users to transact across multiple chains seamlessly. Bitte even has a natural language interface for commands. We also have financial applications, multi-chain DEXs, lending across different assets, and more. On the AI side, crowdsourcing applications like NEAR Crowd have been running for years, improving data acquisition at lower costs. All of these pieces are coming together, and NEAR has been growing exponentially in active users. Someone tweeted that there are more side events at EthCC than blockchain apps. How can the space improve? That's why we're doing AI Developer and the AI Hub. We can try to get people to build applications or make it easy to build them. There are many misaligned incentives in the space, and just talking about them doesn't fix the issue. We've invested in bringing existing applications with users, integrating and interconnecting them, which is why we have the most users in Web3. Building consumer apps is hard, and it's easier to fundraise for infrastructure. We want to simplify the process of building applications so people can experiment more and innovate.
