3 ms·
My bet: 1. Advertisers don't want to be featured on an ISIS recruitment video, or something like that. 2. Google implements a filter. It is fuzzy and based on
by jlouis 9y ago
My bet:
1. Advertisers don't want to be featured on an ISIS recruitment video, or something like that.
2. Google implements a filter. It is fuzzy and based on machine learning, because Google are strong believers of machine learning.
3. The algorithm let a single ISIS recruitment video through. Advertisers form a coalition and yell "NO ISIS RECRUITMENT VIDEOS!"
4. This forces Google to implement either whitelists, or tune the algorithm for detecting bad content such that it produces a lot of false positives.
5. Most people are now demonetized, and since the system is based on machine learning, it is not possible for Google to tell you exactly why the neural network decided what it did.
It is clear that the current demonitization algorithm hits unfairly hard on content creators. It forces them to run a Patreon-model. I'm willing to bet that Youtube is now powerful enough that people are willing to pay for content and no ads for certain channels, so in the end this is not necessarily good for advertisers.
- klodolph 9y ago> It is fuzzy and based on machine learning, because Google are strong believers of machine learning. It seems like YouTube must be backed into a corner where that's the only feasible way to do it, at least as a first pass. Video is uploaded at what, 5000x realtime? Or 100 years of video per week?
- polishTar 9y ago>400 hrs of content are uploaded every minute. You’re exactly correct that an algorithmic first pass is the only thing that’s feasible. It’s important to mention that creators can appeal monitization decisions to a human reviewer for videos with at least a thousand views.
- pavel_lishin 9y agoI'd love to see how many minutes of video-with-1000+-views gets uploaded per minute.