Episode Summary
Executive Summary: This episode explores two converging frontiers: ZKML and the latest ZK proving breakthrough, Nova. Daniel Short explains how zero-knowledge proofs can verify AI/ML inference on-chain, preserving authenticity and security for DApps, games, DeFi, and eventually off-chain systems. Justin Drake then breaks down Nova as a major prover-side optimization that makes ZK rollups cheaper, faster, and more decentralized, with broader implications for blockchain scalability and trustless computation.
Main Topics: ZKML and accountable machine intelligence (Priority: 5/5): Daniel Short frames ZKML as a way to prove that a machine-learning model actually ran correctly and produced a specific output, enabling trustless AI inference on-chain and beyond. Why AI and crypto are converging now (Priority: 5/5): The episode argues that the explosion of AI models plus major progress in ZK systems creates a unique moment where verifiable AI becomes technically and economically viable. Use cases for on-chain AI models (Priority: 4/5): Examples include AI-driven liquidity rebalancing, chess engines, generative art, recommender systems, NPC behavior, and other model-driven game or DeFi logic that must not be secretly swapped or manipulated. Nova and the evolution of SNARK proving (Priority: 5/5): Justin Drake explains Nova as a prover optimization that folds structured computations recursively, delivering roughly 10x better performance for elliptic-curve-based SNARKs. Decentralized proving and rollup liveness (Priority: 5/5): A major theme is how cheaper proving enables decentralized prover networks, improving censorship resistance and liveness so rollups are less dependent on centralized infrastructure like AWS. Broader trustless computing and coprocessors (Priority: 4/5): The conversation extends SNARKs beyond crypto to a future where cloud, mobile, banking, and other external computation can be mathematically verified with coprocessors. Zuzalu as a cross-pollination hub (Priority: 3/5): The episode closes with reflections on Zuzalu as a place where cryptographers, builders, and policymakers mix, surfacing unexpected projects like central-bank-backed stablecoins.
Key Arguments: ZK proofs are valuable because they let users verify that compute was done correctly without redoing the compute themselves. ZKML extends blockchain security guarantees to AI outputs, so a DApp or protocol can trust that a specific model—not a swapped-out or malicious one—produced the result. As AI models become more widespread and powerful, the attack surface for manipulation, bias, and rug pulls grows, making verifiability increasingly important. Nova improves the prover side of SNARKs by folding structured computation, making proof generation materially cheaper and faster. Cheaper proving lowers barriers to decentralized proving, which improves rollup liveness and reduces dependence on centralized provers. The long-term vision is not just scaling crypto, but creating a generalized trustless compute layer for the broader internet.
Data Points: Modulus Labs team size: 4 people - Daniel says Modulus Labs is currently a four-person team. Company age: 7 months - Daniel says the team started about seven months earlier. Nova performance improvement: ~10x faster - Justin says Nova is on the order of 10x faster than previous proof techniques for elliptic-curve-based SNARKs. BLS signature verification time: ~1 millisecond per signature - Justin uses BLS verification as an example of linear verification cost. Example bulk verification: 10,000 signatures in ~10 seconds vs ~1 millisecond folded - Illustrates the benefit of folding/Aggregation versus naive verification. Current prover cost: ~1 cent per transaction - Justin says proving cost today is around one cent per transaction. Redundancy target: 100:1 to 1,000:1 - Justin describes desired redundancy for decentralized prover networks. Validator stake requirement reduction: 1500 ETH to 32 ETH - Justin cites BLS aggregation as enabling Ethereum validator requirements to drop dramatically. Ethereum throughput example: 3 million to 15 million gas every 12 seconds - Justin references current layer-one gas context when discussing scaling limits. Applied cryptographer excitement at Zuzalu: 8/10 - Justin estimates excitement around Nova among applied cryptographers at Zuzalu.
Pivotal Quotes: "accountability or integrity technology" — Daniel Short: Daniel defines what zero-knowledge proofs are doing in the ZKML context. "Snarks are just going to completely change the world." — Justin Drake: Justin explains why proving technology matters far beyond a narrow crypto use case. "Snarks are going to eat the world every time you reduce the prover cost by 10x." — Justin Drake: Justin argues that each major cost reduction opens up a much larger design space for applications.
Implications: ZKML could make AI outputs verifiable and harder to manipulate, while Nova could accelerate rollups and decentralized proving. Together they point toward a future where compute across crypto and beyond becomes more trustworthy, composable, and censorship-resistant.