Why Foundation Models Haven’t Replaced Classical Machine Learning
In this episode, co‑founders Dora Shin and Mustafa Abdel‑Baki of Disarray explain why foundation models haven’t supplanted classical machine‑learning (ML) for enterprise tasks like forecasting, fraud detection, and recommendation. They argue that foundation models struggle with proprietary, tabular data and multimodal integration, while classical models excel at handling structured data at scale. Disarray’s solution builds a semantic knowledge graph that unifies fragmented data, code, and organizational context, enabling agents to automate the full ML pipeline—from data ingestion to production deployment. Their approach emphasizes entity resolution and self‑service connectors to reduce engineering overhead and maintain accuracy.
When "Garbage In, Garbage Out" Gets It Wrong
In this episode, Terence Lee St. John, founder of Enly and lead author of the paper "From Garbage to Gold: A Data Architectural Theory of Predictive Robustness," explains why machine‑learning models can achieve state‑of‑the‑art performance even when trained on noisy,...
As Code Generation Speeds Up, Who Tests the Output?
In this episode, CTO Evan Marshall of Ito AI discusses the growing gap between rapid AI‑driven code generation and the ability to reliably test that code. He explains how Ito AI’s platform, Edo, provides automated, execution‑based QA for every pull...
Why Your AI Committee Might Be Your Biggest AI Problem
In this episode, evangelist Samudis examines why many large enterprises are creating AI committees or governance bodies and how these structures can unintentionally slow AI adoption. He contrasts the traditional Center of Excellence model with newer, often politicized committees that...
Your First AI Employee Is Already Clocking In
In this episode, Kay Ju, co‑founder and CTO of GenSpark AI, explains how GenSpark Claw turns an AI model into a secure, cloud‑hosted "AI employee" that runs in its own virtual machine. He highlights three core advantages: isolated cloud VMs...
Coding Agents Meet Data Science
In this episode, host and guest Mikio Braun discuss the emerging role of coding agents—AI tools that generate code—in data science workflows. They explore how these agents excel at writing code but often lack the skepticism and domain awareness needed...
World Models Are Here—But It’s Still the GPT-2 Phase
In this episode, Jeff Hopp, CTO of Odyssey, explains their frontier world model, Odyssey 2 Pro, which generates continuous, interactive video streams that simulate potential futures from a starting image. He describes how the model is trained on massive public...
Teaching AI How to Forget
In this episode Ben Lorica interviews Ben Luria, CEO and co‑founder of Hirundo, about the rising importance of machine unlearning for enterprise AI systems. They explore how organizations can remove or forget specific data points from trained models to comply...