A new paper on arXiv addresses the challenge of governing short-video content at scale. The authors note that the explosive growth of short videos has made content governance significantly harder, while traditional manual labeling cannot keep up with demand.
The paper argues that manual labeling is inefficient, slow to iterate, and costly. To address this, the authors explore and develop a multi-stage autopilot labeling system, which appears intended to automate much of the labeling pipeline rather than relying on human effort at every step.
The abstract presents this as an exploration and development effort, but does not include evaluation details or performance comparisons. As a result, the work is best understood as a proposed direction for automating short-video content labeling, not yet a proven solution.