In this episode of Exceeding Expectations, host Laurel Rockall is joined by Dan Roth, Chief AI Scientist at Oracle, who explains why language models are powerful but not the reasoning engines most people believe them to be, why the relationship between AI and data runs in both directions, and why most organizations are underinvesting in the data side of that equation.

What You’ll Learn:

Why language models are generation machines and what that means for enterprise deployment
The bidirectional relationship between AI and data that most organizations are only investing in one direction
Why information retrieval is far from solved and how models fail at multi-step queries
What a semantic data layer is and why it’s the highest-leverage AI investment almost no one is making
The concept of visibility of failures and why invisible AI errors are the biggest enterprise risk
How AI use is already eroding critical thinking and why leaders must invest in both technology and human judgment

Dan Roth is Chief AI Scientist at Oracle, bringing decades of experience building enterprise AI systems that actually ship to production. He is one of the rare voices in AI who moves fluently between research and production. Dan previously led the science behind Amazon’s first generative AI products and has published more than 450 papers across natural language understanding, machine learning, and reasoning. His PhD dissertation at Harvard, titled “Learning in Order to Reason,” anticipated ideas the entire field is now converging on three decades later.

Episode Resources:

Dan Roth on Linkedin: https://www.linkedin.com/in/dan-roth-8667361
Oracle Website: https://www.linkedin.com/company/oracle
Laurel Rockall on Linkedin: https://www.linkedin.com/in/laurel-rockall-a73813128
Encore AI website: https://www.gainencore.ai