探索區塊鏈與機器學習的未來

9:00 AM – 9:15 AM

Opening Remarks

Bhaskar Krishnamachari, USC

A warm welcome to kick off the day, setting the vision for a future where decentralized systems and AI drive innovation.

Recorded Session

9:15 AM – 9:45AM

The Future of Decentralized Innovation

Tim Draper, Draper Associates & Draper University

Tim discusses emerging trends and long-term strategies for integrating decentralized technologies into our digital future.

Recorded Session

9:45 AM - 10:15 AM

Empowering One Billion Software Creators

Michele Catasta, Replit

For decades, building software has been accessible only by a small portion of knowledge workers. Large Language Models (LLMs) are empowering everyone to generate code, but building software requires setting up the development environment, orchestrating infrastructure, enforcing security, and more. Replit Agent is the only full-stack software creation agent that marries the power of enterprise-grade infrastructure with the ease of use of a natural language interface approachable even by non-technical users. Replit President, Michele Catasta, talks through the technical challenges and research opportunities of developing one of the most used software creation agents in the world.

Recorded Session

10:15 AM - 10:45 AM

Privacy-Preserving AI Compute with Decentralized Verification

Raluca Ada Popa, Berkeley

With compute resources for AI being scarce, users are increasingly leveraging decentralized systems. However, users need to ensure the confidentiality and integrity of their data on these systems. In this talk, I will overview our approach of privacy-preserving AI computation with decentralized verification. Our techniques combine shared security based on restaking, hardware enclaves, and encrypted RAG retrieval.

Recorded Session

10:45 AM - 11:15 AM

Practical and Extensible Decentralised Identity Management

Salil Kanhere, UNSW

Professor Salil Kanhere (UNSW, Australia) explains Self-Sovereign Identity (SSI) through CredChain, a blockchain platform for secure and selective credential sharing. He also introduces the Verifiable Credential-based Trust Propagation Protocol (VCTP), enabling individuals to act as trusted issuers.

Recorded Session

11:15 AM - 11:45 AM

The Rise of Autonomous Commerce

Christoph Niemann, AWS

This presentation will explore the integration of AI agents with digital assets, focusing on leveraging Amazon Bedrock to create a powerful, private knowledge base for real-time crypto data access and transactions. The presenter will detail the architecture for building an intelligent Web3 agent that understands user intent, automates complex processes, and interacts with blockchain networks. Key technical aspects include the development of AI-driven systems as autonomous economic actors, such as crypto trading bots, smart wallets, self-managing DAOs, and IoT-enabled micropayments. This talk will demonstrate how to enable conversational, automatic trading and personalized insights while maintaining security and reliability in the digital asset landscape. By combining Amazon Bedrock with Web3 technologies, the presentation showcases how AI agents can revolutionize user experience in crypto interactions, making them more intuitive and efficient through natural language commands and automated wallet management.

Recorded Session

11:45 AM - 12:45 PM

Panel Discussion: AI for Blockchain

Moderator:

Anand Iyer, Lightspeed Venture Partners

Panelists:

Ashish Goel, Stanford

Christoph Niemann, AWS

Gilad Buchman Coinbase

This panel will explore the transformative integration of artificial intelligence with blockchain technologies, examining how this convergence is creating new paradigms for decentralized governance, smart contract optimization, and predictive analytics in digital asset management. Our distinguished speakers will discuss how AI-powered systems are enhancing blockchain scalability and security while addressing critical challenges including regulatory compliance, energy consumption, and the ethical implications of autonomous blockchain agents.

Recorded Session

12:45 PM - 1:15 PM

Towards an Agentic Internet

Erik Reppel, Coinbase

The explosion of agents has brought light to a fundamental problem of the internet: the highest value actions are hidden away from the public internet. In a human operated, narrow internet this has been an inconvenience, but now this fact is one of the largest limiters for truly autonomous agents. In his talk Erik will discuss how we get from where we are, to a fully composable, agent friendly internet, and crypto’s important role in making that transition. 

Recorded Session

1:15 - 1:45 PM

The Roadmap to a User-Owned AI Model

Illia Polosukhin, NEAR

“Attention Is All You Need” coauthor and NEAR Protocol co-founder Illia Polosukhin shares a vision for User-Owned AI, or AI whose value function optimizes users' wellbeing and privacy, not corporate profits. User-Owned AI provides a business model for open source AI while letting users and companies retain control of their data. Illia shares how blockchain and TEE together enable this future and shares the roadmap to a state-of-the-art, user-owned model that is transparent, decentralized, and upgradable.

Recorded Session

1:45 PM - 2:15 PM

The API Privacy Paradox: How LLM APIs Unintentionally Reveal Model Details

Xiang Ren, USC and Sahara AI

Model APIs serve as gateways to powerful large language models (LLM) while protecting proprietary details. However, subtle signals in API outputs — such as log probabilities and token rankings — can unintentionally reveal critical model attributes, including hidden layer sizes, parameter structure, and even embedding subspace. This talk presents how adversaries can exploit API outputs to infer these hidden properties and discusses mitigation strategies, from modifying API designs to rethinking model architectures. We also examine whether such leaks pose a security risk or offer opportunities for model verification and accountability. Ultimately, we aim to strike a balance between API transparency, trust, and security in LLM deployment.

Recorded Session

2:15 PM - 3:15 PM

Panel Discussion: Blockchain for AI

Moderator:

Yuga Cohler, Coinbase

Panelists:

Ling Liu GA Tech

Dylan Zhang Pond

Jan Liphardt OpenMind, Stanford

This panel will explore how blockchain technology is transforming artificial intelligence development through decentralized data marketplaces, transparent model governance, and tamper-proof algorithm verification systems. The discussion will highlight smart contract systems that establish verifiable trust in AI decision-making processes, offering attendees insights into this revolutionary convergence that promises to democratize access to advanced artificial intelligence while enhancing security, transparency, and accountability.

Recorded Session

3:15 PM - 3:45 PM

Algorithmic Aspects of Decentralized Exchanges

Ashish Goel, Stanford

Given the decentralized nature of blockchain infrastructure, there has
been great interest in decentralized exchanges for cryptocurrencies.
In this talk, we will discuss some of the algorithmic issues that
arise in the design of these exchanges. We will start by describing
how batch exchanges which simultaneously clear multiple assets can
address issues of low liquidity, and present an algorithmic framework
for pricing and matching in this setting.We will then discuss Constant Function Market Makers (CFMMs), which
provide liquidity between a pair of assets by maintaining a liquidity
pool and doing algorithmic pricing. We present an optimization
framework for optimum design of Constant Function Market Makers
(CFMMs), and show how CFMMs can be integrated as participants in a
batch exchange.

Recorded Session

3:45 PM - 4:15 PM

Lowering Barriers to Building In the Era of Open-source AI

Jasper Zhang, Hyperbolic

Jasper will explore how open-source AI is transforming innovation, as community-driven models and frameworks rapidly close the gap with proprietary systems. He’ll also examine what’s still needed to lower the barriers to building AI—highlighting the importance of scalable global compute networks and verification mechanisms like Proof of Sampling.

Recorded Session

4:15 PM - 4:30 PM

Concluding Remarks

Rajarshi Gupta, Coinbase

Thank you all for joining us—let’s carry today’s insights forward as we build the future of AI and blockchain together.

Recorded Session