Anonymous at Sun, 10 Mar 2024 06:03:06 UTC No. 16066034

>>16066032
>Considering the relative size of the view count compared to the english speaking population who watches youtube videos, it must be very likely that the people who view the video are the people the video is actually about no
probably true, and it IS extremely well-targeted,
but "the group of people the video is about" from the algorithm's perspective could be "people who think they have autism". in fact, this seems more likely to me. if you did have autism, there would be no correlation between that and being interested in autism. however, if you're interested in yourself having autism, or any other quality, you're definitely going to be thinking about autism or that quality more than the average person.
who would result in more ad-viewership (total view time) as a reaction to this video; people with autism, or people who already think about autism? is there any reason to think you're more conscious of your brain's behavior than this perfect product of economically incentivized evolution, even with the more limited data it has?

Anonymous at Sun, 10 Mar 2024 06:09:22 UTC No. 16066038

>>16066034
>from the algorithm's perspective could be "people who think they have autism
I agree, but the smaller the view count, the less likely it is because if they were really the target of the algorithm, then it would be reasonable to think the view count would be bigger since much more people think they have autism than people who actually have it. If you end up on a video that has 30k or 70k views on a topic like autism, when Sabine's video has 1.4 million views, it must be that you are much more likely to actually be autistic. I would say similar estimations can be made about IQ and 4chan board. Without an I test, it's reasonable to think that people who mostly visit 4chan on /sci/ or come on the board regularly must have a higher average IQ than the other boards.

Anonymous at Sun, 10 Mar 2024 06:17:30 UTC No. 16066045

>>16066038
you might be right.
but i don't think 4chan boards vs. these videos are comparable as destinations; you (more or less) choose to go to a 4chan board, while the algorithm might have incentives to catch people in behavior loops. the algorithm doesn't JUST show people the optimal video they would pick if they knew about every video, the algorithm shows people the stimuli which maximize time spent on the platform.
i'm also not sure IQ vs a diagnosis are comparable, but it seems reasonable for these purposes.
i do wish we could read what the algorithms "intentions" are when showing the things it shows. i wonder what sort of insights it could give us.

🗑️ Anonymous at Sun, 10 Mar 2024 06:24:02 UTC No. 16066055

>>16066045
I agree that those are two different contexts but the nuance is that I mentioned the idea is to see it from a bayesian perspective. Even if there was a personal selection on 4chan, the fact that you end up on /sci/ and come back here at least once makes it MORE likely for you to be high IQ, despite not being a necessary reason to think it's the case categorically. Algorithms just make the increase in probability dramatic and the inference can be easily done the the viewer's part by considering the size of thlhe population that speaks tbeir language and the view count alone.

Anonymous at Sun, 10 Mar 2024 06:26:42 UTC No. 16066057

>>16066045
I agree that those are two different contexts but the nuance is that I mentioned the idea is to see it from a bayesian perspective. Even if there was a personal selection on 4chan, the fact that you end up on /sci/ at all and come back here at least once makes it MORE likely for you to be high IQ, despite not being enough of a reason to think it's the case categorically. Algorithms just make the increase in probability dramatic and the inference can be easily done on the part of the viewer by considering the size of the population that speaks their language and the view count of the video alone.

Anonymous at Sun, 10 Mar 2024 11:38:20 UTC No. 16066271

>>16066032
I've had autism/incel videos recommended before and wondered the same. Some of those videos had more than 70k views. Youtube recommends videos based on what others with similar tastes watch. If you're in the same digital room as autistic people you might be autistic too. This isn't a diagnosis on its own- you aren't considering the amount of recommended videos you get that don't resonate with you. Most people getting autism videos will dismiss them like any other bad recommendation, but you've gone on 4chan of all places to write paragraphs about it. You are autistic for that, not for the recommendation.

Anonymous at Sun, 10 Mar 2024 14:24:03 UTC No. 16066403

>>16066032
No

Anonymous at Sun, 10 Mar 2024 15:04:20 UTC No. 16066446

>>16066032
The algorithm does not need to know what kind of video is being selected, only the specific video being selected. The algorithm is going to make its own categories of what videos are selected together by their frequency in the group, not by subject matter or topic.

Here’s how it works: People choose videos they are interested in from the videos presented, and from their interests outside of watching videos, but the algorithm sees only both the videos they choose, AND what videos they choose from the list provided. It does not care about what the video is about, only what video it is.
By watching their choices over time, a set of different videos appears by frequency. By taking the data from all people, overlap in the videos watched appear. This is the first stage. By taking the data of all people, overlap in the videos selected from the “you might also be interested in” list appear. This is the second stage. By offering the videos with the highest probability of both first stage and second stage patterns, the amount of time on the platform is maximized. Then comes the third stage of advertisement offerings on those choices. By seeing what ads are watched more for the first and second stage patterns, ads are attached to those videos.

So here’s my point: at no time does the algorithm have anything to do with the content of the video. The categories of videos you and I would make based on content and subject never enter into the algorithms of the platform. The algorithm makes its own categories based on usage, and are maximized for time on platform and ad revenue, not topical interest.
(cont)

Anonymous at Sun, 10 Mar 2024 15:06:54 UTC No. 16066447

>>16066446
(cont from >>16066446)
So the algorithm has no idea what the subject or topic is.
This is how Meyers Briggs and Astrology work. Take a lot of orthogonal traits. You choose the trait you think you are from a list of many orthogonal traits. If lots of people choose traits, then patterns form in their choices that relate different orthogonal traits.
Instead of some other criteria determining the categories, the choices of the individuals in group determine the categories.
To then claim correlation is circular. You could have created a different correlation simply by selecting different orthogonal traits.