3 ms·

>One thing that frustrates me most is that even if people do reach the point of understanding the power these people have, it still won't matter much. We still

by kongin 5y ago

>One thing that frustrates me most is that even if people do reach the point of understanding the power these people have, it still won't matter much. We still trust online media such as Reddit, Twitter, Facebook, and even this website, as being the fountain of the mythical phenomenon known as "public opinion". I'm much more interested in who the moderators of social media are and who pays them. Going after 'trolls' is to think we still live in 2011 when this was the way to astroturf. Reddit for example, on average has none of the content users post on their site reach and stay on the front page: https://www.reddit.com/r/RedditMinusMods/ https://www.reddit.com/r/RedditMinusMods/ HN is fairly straightforward in moderation: what dang doesn't like gets deleted, but the user flags have also been weaponised recently. Twitter and FB I have no idea about. But I can't imagine it's hard to find out who the paid moderators are and pay them some more to delete content you disagree with and not delete content you agree with. The majority of them are paid peanuts to begin with.

robbedpeter 5y ago
For most big forums, Twitter, reddit, and Facebook, it's probably 2$ an hour workers in 24/7 facilities in India, the Philippines, and other low wage parts of the world. Automation is being used at an accelerating pace, and there's so much data being pumped online that only a vanishingly small percentage of flagged media gets reviewed by an actual human. A dozen or so motivated and clever humans is all it takes to get a person or channel suspended on any of the big platforms because automation is so embedded. Situations that a human world recognize as an error usually get corrected, but almost all social media is too much data by far to include humans at a meaningful scale. Moderation at scale is currently a very blunt instrument, compared to a well curated, civil forum. Transformers and other recent models have enabled very nuanced classification of text, so we're going to see a confluence of economic factors - the quality to cost ratio of humans vs software at all levels of moderation and curation. Eventually the bots will be just barely good enough and cheap enough to run at scale that humans will be removed from the loop, from first pass sorting all the way up the chain. The humans will only set the parameters, and then we'll have a whole new bag of problems. It'll be easier to manipulate software to trigger censorship than to manipulate humans following a set of guidelines. https://news.ycombinator.com/item?id=23918763 https://news.ycombinator.com/item?id=23918763 Because of worker protections, minimum wage, and legal liabilities, it's better to find cheap multilingual offshore moderation farms than to hire in the US. It's a race to the bottom. The problem of moderation at global scale may prove to be intractable without human level ai.

by kongin 5y ago

robbedpeter 5y ago

For most big forums, Twitter, reddit, and Facebook, it's probably 2$ an hour workers in 24/7 facilities in India, the Philippines, and other low wage parts of the world. Automation is being used at an accelerating pace, and there's so much data being pumped online that only a vanishingly small percentage of flagged media gets reviewed by an actual human. A dozen or so motivated and clever humans is all it takes to get a person or channel suspended on any of the big platforms because automation is so embedded. Situations that a human world recognize as an error usually get corrected, but almost all social media is too much data by far to include humans at a meaningful scale. Moderation at scale is currently a very blunt instrument, compared to a well curated, civil forum. Transformers and other recent models have enabled very nuanced classification of text, so we're going to see a confluence of economic factors - the quality to cost ratio of humans vs software at all levels of moderation and curation. Eventually the bots will be just barely good enough and cheap enough to run at scale that humans will be removed from the loop, from first pass sorting all the way up the chain. The humans will only set the parameters, and then we'll have a whole new bag of problems. It'll be easier to manipulate software to trigger censorship than to manipulate humans following a set of guidelines. https://news.ycombinator.com/item?id=23918763 https://news.ycombinator.com/item?id=23918763 Because of worker protections, minimum wage, and legal liabilities, it's better to find cheap multilingual offshore moderation farms than to hire in the US. It's a race to the bottom. The problem of moderation at global scale may prove to be intractable without human level ai.