3 ms·

> 1. Youtube's system is probably not stateless. It most likely looks at the uploaders history and other channel data points. We dont know how Youtube´s system

> 1. Youtube's system is probably not stateless. It most likely looks at the uploaders history and other channel data points. We dont know how Youtube´s system operates which is actually part of the problem as the lack of transparency makes it hard for anyone to make any reasonable assessment. > 2. An upload with no video or audio content is an abnormal data point. It's an outlier and doesn't represent real case scenario. Not really as titles (combined with other meta-data) have for a long time impacted the state of monetization, suppression and view count for any given video regardless of its content.

ehsankia 7y ago
> We dont know how Youtube´s system operates Sure, which is why this isn't "conclusive proof Youtube is blocking LGBTQ content". As I stated above, it's a contrived experiments and the results most likely can't be extrapolated, and you can't reach any definitive conclusions from it. > Not really as titles (combined with other meta-data) have for a long time impacted the state of monetization I'm not saying they don't play a role, most likely all of these data points are fed into the model. What I'm saying is that Youtube is primarily a video/audio site, so running an experiment with no audio/video isn't testing the model the way it runs in practice. The model wasn't made to classify only titles, this is an abnormal input and therefore the output can't be used to reach any conclusions.
- wesammikhail 7y ago
  Your criticism of the testing methodology is fair and valid. However, I see this as a test controlling for a single variable to the best of the ability of whoever made it. I am not sure what a better testing methodology would look like. Which is why I see this at least as an indication of what is systematically going on.

3 ms·

> 1. Youtube's system is probably not stateless. It most likely looks at the uploaders history and other channel data points. We dont know how Youtube´s system

by wesammikhail 7y ago

ehsankia 7y ago
> We dont know how Youtube´s system operates Sure, which is why this isn't "conclusive proof Youtube is blocking LGBTQ content". As I stated above, it's a contrived experiments and the results most likely can't be extrapolated, and you can't reach any definitive conclusions from it. > Not really as titles (combined with other meta-data) have for a long time impacted the state of monetization I'm not saying they don't play a role, most likely all of these data points are fed into the model. What I'm saying is that Youtube is primarily a video/audio site, so running an experiment with no audio/video isn't testing the model the way it runs in practice. The model wasn't made to classify only titles, this is an abnormal input and therefore the output can't be used to reach any conclusions.
- wesammikhail 7y ago
  Your criticism of the testing methodology is fair and valid. However, I see this as a test controlling for a single variable to the best of the ability of whoever made it. I am not sure what a better testing methodology would look like. Which is why I see this at least as an indication of what is systematically going on.