You are viewing a single comment's thread. View all
0
PurestEvil on scored.co
1 year ago0 points(+0/-0)
> Criminalizing speech that criticizes a government.
And israel and jews. Don't forget that detail!
> Using AI to monitor social media and punish those that post wrongthink.
It appears there are already relatively fast algorithms existing that ensure a good amount of censorship without the need for LLM-like AIs. YouTube comment sections take 20 seconds to evaluate if a reply comment gets auto-deleted, and it seems to be keyword based. So if there are words of "negativity" and some ethnicity mentioned, the entire comment gets auto-deleted. Even words like "tribe" are likely in the keywords.
I have a vague sense of how it works, as I'd be able to do it as well. I've already created a web-crawler years ago that had the purpose to pick up all sorts of words related to a topic like "gaming" minus the words that are generic like "and", "if", "then." It involved a lot of float value based calculations and evaluations. I've done these evaluation based algorithms multiple times already, and that's likely how they did the YT algorithm too.
I've also done a name associator algorithm. So in 2 lists of names which are similar but different like "A. Hitler", "Adolf Hitler", "Hitler", "Adolf", "AH", "AHitler", "Dr. Hitler", it found the best matches, and recognized these for example as very similar. It took me 2 weeks, but it was never used.
For YouTube it just took years in teams for them... which would have took me a few weeks or months probably.
And israel and jews. Don't forget that detail!
> Using AI to monitor social media and punish those that post wrongthink.
It appears there are already relatively fast algorithms existing that ensure a good amount of censorship without the need for LLM-like AIs. YouTube comment sections take 20 seconds to evaluate if a reply comment gets auto-deleted, and it seems to be keyword based. So if there are words of "negativity" and some ethnicity mentioned, the entire comment gets auto-deleted. Even words like "tribe" are likely in the keywords.
I have a vague sense of how it works, as I'd be able to do it as well. I've already created a web-crawler years ago that had the purpose to pick up all sorts of words related to a topic like "gaming" minus the words that are generic like "and", "if", "then." It involved a lot of float value based calculations and evaluations. I've done these evaluation based algorithms multiple times already, and that's likely how they did the YT algorithm too.
I've also done a name associator algorithm. So in 2 lists of names which are similar but different like "A. Hitler", "Adolf Hitler", "Hitler", "Adolf", "AH", "AHitler", "Dr. Hitler", it found the best matches, and recognized these for example as very similar. It took me 2 weeks, but it was never used.
For YouTube it just took years in teams for them... which would have took me a few weeks or months probably.