By Mukund Sundararajan, Google, Mountain View, CA, USA

Abstract 

As Artificial Intelligence systems become increasingly capable, a critical paradox emerges: how can a human evaluator verify an answer to a problem they cannot solve themselves? This talk discusses verification protocols for tasks in mathematics, science, general knowledge, and programming. Performing this kind of verification is critical to the effective adoption of AI and is a broadly useful skill.

Bio:
Mukund Sundararajan is the Quality Lead for Gemini.