Skip to content
No PriorsNo Priors

No Priors Ep. 32 | With NEAR’s Illia Polosukhin

More than 25 million users are using NEAR-powered applications. Co-founder of NEAR protocol and Transformers author Illia Polosukhin joins hosts Sarah Guo and Elad Gil to discuss the intersections of crypto and AI technology, what we should expect from AI agents, decentralized data labeling, why AI’s alignment problem is really a human problem, and more. NEAR allows users to effortlessly create and distribute innovative decentralized apps across any blockchain, while helping build a more open web. Before co-founding NEAR, Illia researched AI at Google and co-authored the landmark paper, "Attention Is All You Need." 00:00 - Blockchain, AI, and Web3 Intersection 09:58 - Blockchain and AI 16:07 - Blockchain and AI Integration Challenges 23:35 - Inference and Decentralized Data Labeling 30:13 - Web 3 and AI SaaS Challenges 38:18 - The Future of Hardware Accelerators

Sarah GuohostIllia PolosukhinguestElad Gilhost
Sep 14, 202342mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Transformers, NEAR, and AI Agents: Building a Verified, Decentralized Web3 Future

  1. Illia Polosukhin, co-author of the Transformers paper and co-founder of NEAR, discusses how his AI background led to building NEAR as a "blockchain operating system" that abstracts Web3 complexity for users and developers.
  2. He explores intersections of AI and blockchain, especially AI agents as economic actors, decentralized data and compute marketplaces, and blockchain-based identity and reputation for content authenticity.
  3. Polosukhin argues that the real challenge is "human alignment"—building societal and technical systems resilient to misinformation at AI scale—rather than just AI alignment in isolation.
  4. The conversation also covers decentralized inference, marketplaces for data labeling, the future of SaaS as database-plus-dynamic-UI powered by agents, and why Transformers are likely to remain the dominant model architecture given current hardware lock-in.

IDEAS WORTH REMEMBERING

5 ideas

AI agents will become economic actors if given blockchain accounts.

Equipping language-model-based agents with on-chain identities and wallets lets them autonomously pay humans and other AIs, manage workflows, and even function as organizational "CEOs" tasked with goals and KPIs.

Blockchain provides critical primitives for content authenticity and provenance.

Cryptographic signing from secure hardware (e.g., phones, cameras) through the processing pipeline, anchored on-chain, can create verifiable trails for media, enabling browser-level indicators (like SSL’s green lock) that content is authenticated.

Human alignment and societal resilience to misinformation matter more than abstract AI alignment.

AI mainly scales pre-existing human problems like propaganda, fraud, and manipulation; robust identity, reputation graphs, and trust frameworks are needed to handle personalized, large-scale misinformation rather than focusing solely on aligning models in the abstract.

Decentralized inference is more practical and valuable than decentralized training.

Massive bandwidth requirements make distributed training across heterogeneous nodes impractical today, while inference requires far more total compute, has strong privacy needs, and can better leverage crypto tools like MPC and zero-knowledge proofs.

Open, tokenized data-labeling marketplaces can improve scalability and fairness.

Web3-based marketplaces allow global workers to participate without local subsidiaries, introduce economically meaningful quality controls (e.g., honeypots and stake-based penalties), and protect annotators from unilateral non-payment common on centralized platforms.

WORDS WORTH SAVING

5 quotes

This is the first time that a machine is able to communicate with people in the same way.

— Illia Polosukhin

I have this view that we need human alignment instead of AI alignment.

— Illia Polosukhin

You can have organizations that are run completely by AI, where the CEO role is taken by an AI agent.

— Illia Polosukhin

We need to build a society that is actually able to deal with effective misinformation at scale.

— Illia Polosukhin

Right now what SaaS is, is one database with a specific UI for a specific problem.

— Illia Polosukhin

Origins of the Transformers paper and Illia’s transition from AI to founding NEARNEAR as a blockchain operating system and its user-facing abstraction of Web3AI–blockchain convergence: economic AI agents, DAOs, and on-chain coordinationContent authenticity, identity, and reputation systems to combat AI-driven misinformationBlockchain-based identity models and why wallet/key UX must evolveDecentralized AI infrastructure: GPU marketplaces, training vs. inference, and privacyFuture of SaaS, agent-driven workflows, and the durability of Transformer architectures

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome