Why The YouTube Algorithm Will Always Be A Mystery

9 insights saved from this video by @technology

@technology
Watch-time optimization cannot recognize truth, quality, context, or safety. Clicks can therefore reward harmful and conspiratorial videos.
@technology
Human moderation after complaints can preserve legal protections. Human approval of everything can make YouTube legally responsible as a publisher.
@technology
Optimizing YouTube for shares suppresses private topics and loyal niche audiences that rarely share anything publicly.
@technology
Neural networks excel when games have obvious scores. Recommendations must first decide what “winning” means for human behavior.
@technology
YouTube’s ideal black box would optimize ad revenue while judging social, financial, reputational, child-safety, and truth-related consequences.
@technology
YouTube cannot inspect or repair an opaque recommendation model. It can only change the feedback signals guiding it.
@technology
A recommendation system can boost watch time immediately while promoting harmful content and creating long-term risks for advertisers.
@technology
YouTube initially gave its recommendation algorithm a single goal: increase watch time.
@technology
Creators want YouTube to recommend their videos even when audiences do not want to watch them.

Keep what you found in Why The YouTube Algorithm Will Always Be A Mystery

A free account saves any of these cards to your Collection, and Korva resurfaces them so you actually remember.