AI + a16z How Foundation Models Evolved: A PhD Journey Through AI's Breakthrough Era
65 snips
Nov 18, 2025 
Omar Khattab, an Assistant Professor at MIT and creator of DSPy, discusses the evolution of AI and the need for programmable intelligence over AGI. He explains how his work aims to bridge the gap between human intent and AI capabilities, emphasizing that a programming language for intent could transform AI from unpredictable models into reliable systems. Omar introduces concepts like DSPy's signature abstraction and the importance of separating intent from models, which could revolutionize how we build and interact with AI.
AI Snips
Chapters
Transcript
Episode notes
Systems Over Raw Intelligence
- Intelligence alone isn't the goal; people want specific, reliable systems that solve concrete problems.
- Models are tools inside systems, and system design matters more than raw model size.
PhD Timing Fueled System Work
- Omar recounts starting his PhD in 2019 as foundation models emerged and methods were immature.
- He focused on how to build systems and applications using these nascent models.
Specification Is The Hard Part
- The key challenge is specifying human intent, not just improving model capabilities.
- Natural language is too ambiguous and code is too rigid, so we need an intermediate abstraction.