The Delphi Podcast Dylan Zhang: Crypto Native AI Models Behind the Future of AI Agents
Nov 18, 2024
Dylan Zhang, co-founder of Pond AI, dives into the transformative blend of AI and crypto, focusing on decentralized finance (DeFi). He explains how Graph Neural Networks (GNNs) outperform Large Language Models (LLMs) in blockchain applications. The discussion covers innovative dynamic fee structures, achieving 92% accuracy in detecting malicious behaviors, and the evolution towards decentralized AI models. Dylan emphasizes model ownership in crypto and the exciting future of AI agents, showcasing the potential for collaborative development in this space.
AI Snips
Chapters
Transcript
Episode notes
Decentralized Model Layer
- Pond AI shifted from a large GNN to a decentralized model layer for broader application.
- This approach allows developers to access various smaller, specialized models and contribute to the ecosystem.
Predictive Accuracy
- Pond AI achieved 92% accuracy detecting malicious behavior by leveraging on-chain data's simplicity.
- This high accuracy stems from the relatively simple and predictable nature of on-chain interactions.
DeFi Implementations
- Pond AI sees a combination of enhancing existing DeFi primitives and building entirely new ones.
- The approach depends on the project's needs and understanding of machine learning's potential.