The Bigger Model Myth | Vanja Josifovski (CEO, Kumo AI)
Listen to full episode:
Summary: Anant and Ed open by reacting to recent turbulence around Anthropic's Fable models, using it to dig into model governance — how enterprises pick, trust, and manage dependencies on AI models in a landscape that keeps shifting. The conversation then turns to guest Vanja Josifovski (Co-Founder & CEO of Kumo AI), a longtime relational database expert, who argues that text and images won AI's early wins while relational data has lagged behind because it's harder to translate into model-ready features. The throughline is that enterprises need to stop chasing headlines and start understanding their own data well enough to make AI adoption durable rather than reactive.
Chapters:
00:00 Introduction and Context Setting
02:48 AI Safety and Model Governance
05:51 Insights from Hard Fork Live
08:57 The Evolution of LLMs and Modalities
11:43 Vanja Josifovski's Journey and Background
14:38 The Evolution of Relational Databases
17:50 Challenges in AI and Machine Learning with Relational Data
20:53 Kumo's Approach to Graph-Based Models
23:51 Understanding Graph Transformers and Their Applications
26:38 The Complexity of Temporal Graphs
29:36 Data Sources and Building Models
32:11 Synthetic Data: The Future of Machine Learning
35:43 In-Context Learning: A New Approach to Model Training
40:36 The Evolution of Predictive Models in Enterprises
44:51 AI Enterprises: The Role of Agents and Neural Networks
49:30 Kumo: Bridging the Gap Between Data and Decisions
52:04 The Future of AI and Data Integration
Sound Bites:
“The model layer is not destined to be a utility.”
“Structured data didn’t lose. It was waiting for the right representation.”
“Synthetic data is not a workaround, it is the unlock.”
“Data locality matters more than model glamour.”
