Design Harmful Content Detection

Frame the Problem: What Makes It Its Own Design

"Design a system to detect harmful content." That covers hate speech, harassment, spam, graphic violence, misinformation, the whole list.

This question combines classification, safety policy, large-scale serving, and human review.

It shares the fraud lesson's adaptive attackers and unequal error costs. This lesson focuses on three additional requirements, although some also occur in fraud systems.

**1. It is multimodal