Inside NEAR: 10+ Years Building Toward the Agent Economy with Illia Polosukhin

**Robbie** (0:00)
Guys, welcome back to The Rollup. Today's episode is with NEAR Protocol founder, Illia, talking about the state of NEAR. This is a quarterly series decrypting and uncovering the state of the NEAR Protocol ecosystem from the intent side, from the AI side, and everything in between. Illia, it's great to have you.
And I know it's been a crazy month, crazy week specifically for you, but the market is clearly responding to something at NEAR right now. Good to have you back, Ben.

**Illia Polosukhin** (0:31)
Thanks for having me.

**Andy** (0:32)
Absolutely. I think there's a lot of misconceptions as well with respect to the NEAR Protocol, your roots in AI, your contributions outside of crypto, as well as the duration of which you've been building this project. So this episode today is Inside NEAR, 10 plus years building toward the agent economy.
Obviously, there's a lot of fluctuations in the market as of late. We're recording this on Friday, June 5th. So this entire week, there's been up, down, sideways and more. But there's something happening beneath the surface. Maybe you could just help us understand in your perspective, what has actually changed and what is important to the market? What are they recognizing beyond just the price? What has actually changed? What is most important?

**Illia Polosukhin** (1:26)
Yeah. I mean, as you mentioned, we've been on this journey for quite a while. The origins of NEAR from 2017 We've been building blockchain components from 2018, went live in 2020
I think the big change that's happening now is this transition from the tech and architecture build out to really adoption and kind of usage focus. And as part of this is also, I mean, effectively flipping the switch on revenue, kind of enabling all the species really to work with each other in cohesion.

**Andy** (2:14)
Yeah, so these things... Yeah, please.

**Illia Polosukhin** (2:17)
No, and kind of starting to see all of these pieces really being useful for people versus being kind of maybe some more conceptual.

**Andy** (2:29)
Yeah, and so I think as of recently, because a lot of these things are sort of starting to gel together and reach a confluence, NEAR has attracted a lot of new interest, a lot of new attention. And there might be some people that have joined the community or the network recently or even used NEAR's products like Iron Claw or NEAR Intense, that may not be fully aware or familiar with the origin story, right? They may not realize your roots in AI, your roots with the Transformer. Can you catch us up and maybe rewind the clock a little bit on the chronology of how these things happened, when you started NEAR, when you contributed and wrote the Transformer paper and your history at Google. Can you just put those things in sequence for us?

**Illia Polosukhin** (3:14)
For sure, yeah. So rolling back all the way to when I was 10 years old, I was really excited about this idea of artificial intelligence, went and read all the science fiction I can get my hands on, from Asimov to more modern things. And even when in high school, I was trying to build first neural networks, which I didn't actually know linear algebra yet, so it was a bit hard to do that.
I actually, while I was still in Ukraine, I was taking all of the first massive online courses, right from Andrew Ng, from Peter Norvig, so I was trying to get all the machine learning knowledge I could. And I joined Google when I saw this idea of, I call it a cat, well, it was called cat neuron, and effectively it was the first example where a neural network learns some concept that is like reasonable to us, like a cat, completely without any supervision, without any human saying, this is a cat, but just showing lots of images, neural network learned this is a cat, right? That was the first example, this is like 2012, 2013, and I'm like, we should do that on text. Text is how we express knowledge. Question answering is how we actually understand if the other person knows and can think about something. And so I joined Google to work on question answering. And Google is the best place for that because, well, there's a billion people asking questions all the time. Right, so there's a ton of data on this, there's a ton of kind of just general impact you can do by solving that problem. And we worked kind of on what were the state of the art neural networks at the time, and the problem was they were too slow. Right, if you put them into production on google.com, they, you know, as human, if you give it 10 pages of text, it will take some time for it to read it before it can actually respond to a question. And that's not how google works, right? Google, you get an, you want an answer at superhuman speed. You want like in 300 milliseconds, it should scan all the internet and give you the answer. And so that's really what gave Burr's kind of ideologically to Transformers. How do we actually consume all of the context in parallel, use kind of parallel computing we have, and give answers, translate, or whatever the task we try to do. And so it was kind of merge of a bunch of ideas. My kind of director at the time, Jakob Skorit, came up with like a concept.