>>104559686
>>104559955
Youtube (and Gmail and everything google) lost control of their "anti spam" software long ago.
It used to be top of the line, based on heuristics but somehow mid 2020 or 2021 they changed it to what I believe to have been the "proto LLM" that today is known as "AI" and instead of having to program the anti spam with code, they just would go
>filter content that looks spurious
>let less nudity to pass through
>flag users that looks suspicious
in plain language, add to a big list of "prompts" and the AI would figure it out who and what to filter or let pass.
Right now I doubt they have any clue of why, once in a while, a huge segment of good content / good audience is filtered while bad content / spam gets through and they can't fix it because adding more prompts or removing existing ones doesn't result in a deterministic outcome.
X faced that exact conundrum recently: after accusations that Grok had turned woke the engineers added a single line to its system prompt, something like (in plain English)
>don't shy out of using alternative viewpoints even if it doesn't come from the sources considered reputable as long as they can be verified true
and then MECHAHITLER was born with Grok going to the right of David Duke and Oswald Mosley.
Google has nobody that can possibly figure out why the viewership dropped in August because it was more likely than not something added to the system prompt of their anti spam viewers
>let fewer people with Adblock pass through
and it filtered out 90% of desktop users
4chan Search
2 results for "6cbad725d88c46b04e8fa12c786d6b56"
>>104199380
>something might be going on at google.
That something is the same thing that started in 2020 with the "2020 algo change": they fully delegated the detection of "spam viewers", from the heuristic deterministic system they used (same as Gmail anti spam btw), to the newfangled
>AI
they're cooking since more or less the early 2020s. Here for instance is people complaining about the same thing, except about their email spam system instead of their "viewer SPAM" system
https://support.google.com/mail/thread/98292364/spam-filter-is-excessively-aggressive-and-no-longer-trainable?hl=en
More or less around the end of 2020 they completely stopped "programming" their anti spam system and started "training" it, first with neural networks and now with LLM but the end result is the same
>whatever result they achieve CANNOT BE FIXED by programmers
all they can do is to "plain text" determine what they consider to be a spam email, a spam viewer and the "AI" system will try to filter using those instructions.
It's pretty much "dead internet theory" coupled with an overzealous AI SPAM system filtering out human content but letting all sorts of automated or AI generated content through.
>something might be going on at google.
That something is the same thing that started in 2020 with the "2020 algo change": they fully delegated the detection of "spam viewers", from the heuristic deterministic system they used (same as Gmail anti spam btw), to the newfangled
>AI
they're cooking since more or less the early 2020s. Here for instance is people complaining about the same thing, except about their email spam system instead of their "viewer SPAM" system
https://support.google.com/mail/thread/98292364/spam-filter-is-excessively-aggressive-and-no-longer-trainable?hl=en
More or less around the end of 2020 they completely stopped "programming" their anti spam system and started "training" it, first with neural networks and now with LLM but the end result is the same
>whatever result they achieve CANNOT BE FIXED by programmers
all they can do is to "plain text" determine what they consider to be a spam email, a spam viewer and the "AI" system will try to filter using those instructions.
It's pretty much "dead internet theory" coupled with an overzealous AI SPAM system filtering out human content but letting all sorts of automated or AI generated content through.