Post by TELUS Digital

A global video-sharing platform with billions of monthly users needed its content-moderation model to catch more than obvious keyword violations. Nuanced signals, evasive formats like memes and coded language, and missing context were slipping past keyword-based detection. TELUS Digital's safety analysts built a three-tier chain-of-thought framework applied across five structured sections per review. Analysts supervised the model's reasoning, not just its outputs, tracing each verdict to a specific policy and addressing exceptions directly. The result was more than 850 complex chain-of-thought evaluations and policy analysis accuracy above 95%, with verdicts specific enough to hold up on appeal. See what it takes to catch what keywords miss in the full case study: https://lnkd.in/gxkeNkZi #TrustAndSafety #ContentModeration #AI