August 24, 2023

Why is one of the founding fathers of generative AI all in on Web3?

Illia Polosukhin created the technology that powers generative AI. Now he's building a Web3 platform to try and transform the internet, again

Tim Smith

6 min read

Illia Polosukhin has a track record of shaping the future of technology. As a member of the team that published the landmark 2017 AI paper 'Attention is All you Need', he’s one of the co-creators of the transformer model — the type of AI architecture that powers generative AI systems like ChatGPT (the “T” stands for transformer).

Polosukhin was one of a team of eight Google staffers who worked on the breakthrough. But, just months after publishing, he left Google to launch NEAR — a startup building an operating system and infrastructure that allows developers to easily build apps in Web3. This refers to a vision for the next generation of the internet — one in which control of platforms is more decentralised and less held by a small number of big companies.

In doing so, Polosukhin crossed over from what has become today’s best-beloved sector by tech investors, to one that’s now commonly derided as the latest fad to be thrown on the trash heap of misguided VC bets.


Sifted sat down with Polosukhin to ask why investors have been misguided to write off Web3, and how the hype around generative AI could be creating a bubble.

Lessons in hype

Polosukhin couldn’t have dreamed about how successful the transformer model has become in AI, but he says that he is taken aback by how many startups and investors are getting excited about so-called “ChatGPT wrapper” companies (ones that build a product layer on top of models like OpenAI’s GPT-4).

“I'm surprised by how people are not thinking one step ahead. You had this wave of startups that were like, ‘Hey, we’re gonna wrap up OpenAI’s API and build X for Y,’” he says. “People got really excited by that, but if you just think one step ahead, the next model from OpenAI is probably just going to be able to do that.”

Polosukhin also thinks we’ve got a long way to go in working out what kinds of products make sense to build with generative AI, with most new ideas building off the chat interface made so popular by ChatGPT.

“People were like, ‘Oh, we can just make this chat user interface for something else.’ But it's not how people want to interact with most stuff,” he says. “There’s no moat. You’re pretty much feeding all your training data to OpenAI — so that their stuff gets better as well — and you haven’t done user experience testing to understand if this is a better experience for that problem, which for most problems, it is not.”

Polosukhin believes that there will be a big role for blockchain as we begin to see a mass proliferation of AI-generated deepfakes and synthetic images and videos across the internet, by providing a more trusted proof of where content has come from.

“What we need is a cryptographic way to trace who created this content, where’s the origin of it?” he says. “We’re building tooling for digital reputation and identity because you need that as a way to create the backbone of a new internet.”

Web3 is already useful

Even if investor interest for Web3 — the hype sector before AI — has waned, Polosukhin is still bullish on a blockchain-based, decentralised vision for digital businesses.  

“Web3 was a regular hype cycle. Things got really overheated, everybody got too excited and then results weren’t immediate — as they never are — and so, people went for the next shiny thing,” he says. “But from the perspective of the technology, there are already things that work really well.”

One of those use cases actually explains how NEAR — which has now raised more than $520m from investors including Tiger Global and Andreessen Horowitz — became a Web3 company in the first place.


The startup began life as an AI business, Polosukhin tells Sifted, and was paying students around the world to do data labelling: the job of categorising raw or unstructured data. Web3, he says, became a much better way of paying these workers who were often doing the work piecemeal, receiving small payments at a time.

“We had people around the world, mostly students from China, Russia, Ukraine, etc., who were doing kind of small tasks for us. Paying these people was actually a pretty painful process — sending money from the US to China is complicated, Russia and Ukraine as well,” says Polosukhin. “We ended up looking at blockchain as a solution for our own problem.”

He adds that payments via blockchain are much cheaper than other alternative providers like Stripe for paying these “micropayments.”

“We actually had a startup just switch over from Stripe to NEAR because it's about 100 times cheaper for microtransactions to use NEAR than Stripe,” he says. “The fee there can be about 25 cents that Stripe charges plus around 2.7%, versus NEAR’s fraction of a cent per transaction.”

Sifted reached out to Stripe for comment and a spokesperson directed us to the company’s pricing page (fees vary depending on geography but are roughly in line with Polosukhin’s estimates in many countries), adding that it offers “different pricing models for larger users.”

The NEAR founder says that the platform currently has around a million monthly active users using its distributed work and payments platform, and will soon be spinning it out into its own company.

Making Web3 great again

Polosukhin says that, while a work and payments marketplace like this can create a solid business model, it’s not the kind of “shiny thing” that people in the tech industry like to get excited by. He does believe, though, that Web3 is still set to trigger bigger disruptions in fields like gaming and social media.

“There's potentially new revenue streams for the game developers. There's community-generated content where the community creators and curators are receiving a portion of the revenue,” says Polosukhin. “You’ll have all these different new ways of interacting with content.”

He admits that it’s still too early to see exactly how people will interact with platforms in Web3, but his goal is to make it as easy as possible for people that aren’t blockchain acolytes to build apps and products with the technology.

“There's lots of blockchains, ‘layer twos’, wallets — you have to understand all those key phrases — so we’re really working on simplifying all that and making it extremely easy to sign up,” says Polosukhin.

He adds that, to help Web3 attract widespread use, NEAR is working on making it easier to move money in and out of traditional bank accounts, as well as building in important compliance systems like know-your-customer tools.

Polosukhin’s vision of Web3 seems to be less about VR goggles or the promise that our lives will be increasingly lived in dystopian digital nightmares built by Mark Zuckerberg, and more about an internet where value is shared more evenly and information is harder to manipulate.

Maybe we shouldn’t give up on Web3 just yet.

Tim Smith

Tim Smith is a senior reporter at Sifted. He covers deeptech and all things taboo, and produces Startup Europe — The Sifted Podcast . Follow him on Twitter and LinkedIn