The Data Exchange with Ben Lorica

Steve Wilson, Chief AI and Product Officer at Exabeam, dives into the complexities of securing Large Language Models and agent workflows. He highlights the unique risks of prompt injection and supply chain vulnerabilities that arise with democratized AI tools. Wilson discusses the importance of guardrails, the dangers of excessive agent authority, and lessons learned from web security mishaps. He also explores the concept of citizen developers and advocates for the OWASP GenAI Security Project to provide rapid community-driven guidance for safer AI practices.

Evangelos Simoudis, a venture investor and corporate innovation expert at Synapse Partners, joins to discuss the dual-edged sword of AI in the workforce. They examine AI-driven layoffs, emphasizing the importance of investment in upskilling and R&D. The conversation highlights how trust in AI outputs can be fragile due to unpredictable behaviors and the need for better government coordination on AI access. They also delve into legal complexities surrounding platform liability as AI systems take on more editorial roles in content creation.

Stefania Druga, an independent researcher and former Google DeepMind scientist, delves into creating AI tools for young learners. She shares insights on how children's natural curiosity informs better AI design. Stefania champions the Socratic method for teaching, highlighting her work on Cognimates as a supportive learning copilot. They discuss the importance of multimodal interfaces and the challenges of current AI education tools. Listen in as she reveals how real-time apps like MathMind address misconceptions in math, pushing for innovative solutions in AI education.

Jure Leskovec, a Stanford professor and co-founder of Kumo.ai, dives into the transformative power of relational foundation models for structured enterprise data. He challenges the current limitations of AI in handling relational data, emphasizing the shortcomings of treating tabular data as text. Jure outlines Kumo’s rapid predictive SQL-like language, innovative graph representations, and the model's ability to handle messy data effectively. He also discusses real-world successes like DoorDash's significant improvements and the potential applications of these models across various industries.

Philip Rathle, CTO of Neo4j and a leading expert in graph technologies, explores the integration of knowledge graphs in enterprise AI. He discusses the real-world application of GraphRAG, detailing how it enhances context for AI agents. Rathle highlights successful enterprises using this technology and warns against overly complex projects. He also showcases tools like the LLM Graph Builder for building starter knowledge graphs and emphasizes the need for clear governance and determinism in AI systems, ultimately illustrating how graphs can significantly improve AI reasoning.

Emmanuel Ameisen, an interpretability researcher at Anthropic and author, dives into the workings of large language models. He explains how these models can resemble biological systems and reveals surprising problem-solving patterns, like predicting multiple tokens at once. Emmanuel also addresses the misleading nature of reasoning outputs and the neural mechanics behind hallucinations. He emphasizes the importance of model calibration, debugging tools, and even shares practical advice for developers. It's a fascinating look at the complexity of AI behavior!

Evangelos Simoudis, a venture investor and corporate innovation expert at Synapse Partners, joins the discussion to unpack how AI is reshaping the workforce and corporate strategies. He explains the complexities behind AI-driven layoffs, categorizing them into untrainability and automation. They also explore the need for ROI in massive capital investments in AI, while discussing how LLMOps can manage financial efficiencies. Evangelos emphasizes the importance of cross-functional teams for successful AI integration, highlighting the evolving landscape of AI as a utility.

Ciro Greco, Co-founder and CEO of Bauplan, discusses revolutionizing data engineering by applying software principles like version control and transactional pipelines to data lakes. He highlights the unique challenges of data work, such as scale and fragmentation, and introduces a git-like branching model for enhanced reproducibility. Ciro emphasizes the importance of transactional guarantees, especially for automated agents, and advocates for a code-first approach to enable safe and efficient interactions with data platforms.

Mike Freedman and Ajay Kulkarni, co-founders of Tiger Data, discuss their innovative creation, Agentic Postgres, a database tailored for AI agents. They explain how traditional Postgres struggles with search and scalability when used by agents and introduce concepts like Fluid Storage for instant database forks. The conversation covers the importance of developer-friendly features and the unique capabilities of their MCP server, which enhances model training while ensuring security. They also touch on using Postgres for unstructured data and the future of database tooling.

Nick Schrock, CTO and founder of Dagster, explores the transformative power of data orchestration in the age of AI. He introduces Compass, a Slack-native tool designed for collaborative data analysis, aiming to replace inefficient ad-hoc dashboards. Schrock explains the concept of context pipelines, crucial for AI strategies, and highlights how Compass improves user interaction and onboard processes. He also discusses the synergy of automation and human contributions, making data workflows more efficient and integrated in team environments.