Detecting AI Generated Content in the Real World
Date:
Abstract
In this talk I discuss the difficulties in applying AI detectors to real world high-stakes scenarios. I discuss the stylistic components that make up AI’s style and how despite being able to detect these traces we are a long way from being able to robustly deploy these systems in any meaningful capacity. I finish by discussing how we may begin to use alternative signals to help aid AI detection when confidence is low.
Location
This talk was given on May 9th, 2025 at Drexel University in Philadelphia as a guest lecture for the Artificial Intelligence class.