Episode Summary
Executive Summary: Laura Shin hosts a discussion with Near co-founder Ilya Polosukhin and Poolside founder Jason Warner on how AI and crypto intersect. They argue blockchain can add provenance, identity, reputation, and resource marketplaces to AI, while AI can improve crypto trading, auditing, and DeFi operations. They also warn that generative AI will intensify misinformation and bot-driven attacks, making cryptographic verification and transparent governance increasingly important.
Main Topics: What AI means and why large language models matter (Priority: 5/5): The guests frame AI as the evolving set of machine-learning capabilities once considered magical. They emphasize that LLMs are the newest major wave because they let humans talk directly to computers, unlike earlier systems that required specialized interpretation. Blockchain as infrastructure for AI provenance and resource markets (Priority: 5/5): Ilya argues blockchain can create marketplaces for data, compute, model components, and crowdsourced labor, improving liquidity and global payments. Jason adds that permissionless, trustless systems can make AI inputs auditable and traceable. AI, crypto, and misinformation/authentication (Priority: 5/5): The conversation focuses heavily on how generative AI can scale misinformation and synthetic content. The speakers propose cryptographic signatures, on-chain provenance, identity, and reputation systems to verify authorship and authenticity. AI for coding, smart contracts, and security (Priority: 4/5): They discuss AI coding assistants, Solidity auditing, Rust smart-contract generation, and future AI systems that can detect insecure or previously unseen code patterns. AI is presented as a tool for both development and defense. Trading, DeFi, and predictive analytics (Priority: 4/5): The guests explain that machine learning already powers trading bots, but LLMs lower the barrier to building trading and DeFi strategies through natural language interfaces. They also note AI can help detect hacks and economic attacks in real time. DAOs coordinating AI and AI coordinating DAOs (Priority: 4/5): They explore AI-run DAOs that assign tasks, manage treasuries, and coordinate contributors under community-defined KPIs, as well as DAOs governing model behavior, data sourcing, and bias correction. Regulation, transparency, and societal safeguards (Priority: 4/5): The speakers criticize performative regulation and argue that regulators should adopt blockchain-based recordkeeping, attestations, and auditability. They believe AI and crypto may ultimately help regulate each other through public, verifiable systems.
Key Arguments: AI is best understood as machine learning becoming more powerful and user-facing; LLMs are important because they let anyone interact with computers in natural language. Blockchain can serve as a coordination layer for AI by enabling global payments, crowdsourced work, and decentralized markets for compute, data, and models. Provenance matters more than simply detecting synthetic content: on-chain signatures can show who authorized content and what data influenced it. AI-generated misinformation is really a human problem amplified by new tools; blockchain can help by attaching identity, reputation, and verifiable records to content. AI already improves code generation and auditing, and this will expand to finding vulnerabilities in Solidity, Rust, and other smart-contract systems. Machine learning has long powered trading bots; LLMs mainly lower the barrier to expressing strategies and interacting with DeFi systems in natural language. AI can help detect hacks and economic manipulation by analyzing time series, transaction behavior, and testnet signals before attacks reach mainnet. DAO structures could be extended so AI coordinates tasks and treasury decisions while token holders set goals and monitor performance. Conversely, DAOs could help govern AI by setting data sources, fine-tuning priorities, and community standards for safety and bias. Regulation should focus on verifiable facts, provenance, and accountability rather than arbitrary technical thresholds such as teraflops limits.
Data Points: Podcast date: July 11, 2023 - Episode date stated at the beginning of the show. Token2049 attendees: over 10,000 - Promotion for Token 2049 Singapore. Token2049 speakers: 200+ - Event promotion mentioned a large speaker lineup. Token2049 discount: 65% off - Promo code Unchained for tickets. Ilya’s machine-learning background: about 10 years - He described working in machine learning before starting startups. Google Research work: TensorFlow and Attention Is All You Need - Ilya said he worked on TensorFlow and the transformer paper that underpins modern LLMs. Google AI product surfaces: google.com, Google Play - Ilya said his research launched on consumer products. GPU cluster cost example: $10 million - Ilya said accessing a GPU cluster often requires at least this level of commitment. Blockchain consensus assumption: up to one-third malicious validators - Jason and Ilya referenced Byzantine fault tolerance in blockchain consensus. DeFi hacks in 2022: more than $3 billion - Laura cited losses to DeFi hacks as motivation for better security tools. Crypto/AI regulation example: more than this much teraflops - Ilya criticized proposed AI regulations based on compute thresholds. Crypto.com promo: zero credit card fees for first seven days - Sponsor offer mentioned in the episode. Arbitrum ecosystem: over 500 projects - Sponsor copy described the size of Arbitrum One + Nova. Crypto.com cashback: up to 5% - Sponsor promo for Visa card rewards.
Pivotal Quotes: "If it works right now, it's machine learning, and if you're raising money, it's AI." — Jason Warner: A joke defining how the industry labels ML versus AI. "The system doesn't just need an update, it needs a complete rewrite." — Sponsor copy: Used in multiple ad reads to frame Web3's value proposition. "You could have a podcast fully generated... next year." — Jason Warner: He warned that synthetic content will scale rapidly and become commonplace.
Implications: The episode suggests AI and crypto will converge around trust, provenance, identity, and automation. Expect more AI-assisted code/security tools, on-chain verification of content and data, and DAO-based coordination—alongside stronger threats from synthetic misinformation and bot-driven attacks.