5 ms·

I think this is a point that tech folks like to sweep under the rug all too often. All of these sound wrong: - Walmarts are huge! It's impossible to build fir

I think this is a point that tech folks like to sweep under the rug all too often. All of these sound wrong: - Walmarts are huge! It's impossible to build fire suppression at that scale. - Ford builds an insane number of cars every year, it's not possible to ensure each car is safe at that scale. - The United States burns 20,000 barrels of oil PER DAY. It's simply not possible to ensure a "safe" product at that scale. - This buffet serves 20,000 plates per day, it's impossible to ensure food safety at that scale. Each of those problems has been ~solved. Or those business wouldn't be operating. You only need to moderate a million messages if your customers are posting that many. If you can't afford to hire moderators to handle those million messages maybe it's your business model that's the problem, not moderation.

mytherin 6y ago
Fire suppression systems fail, cars get recalled or malfunction, people get food poisoning in restaurants or from food bought at a supermarket. Any product, not just tech, runs into problems when manufactured at scale. Tech is more visible because it is newer and in the spotlight. Nobody says "stop manufacturing cars" when one manufacturer issues a safety recall because cars have been around so long. The entire premise of producing 100X fewer cars so we can hand-craft cars and hand-check them is ridiculous - we simply accept the occasional manufacturing defect or product recall as the rates at which they happen are so small. These are all rounding errors, and the benefits of scale far outweigh the cons. The same applies to tech. Content moderation has been ~solved as well. The overwhelming majority of content on social media platforms, websites, etc is regular, legal content. The illegal content that makes it through the filters and stays up for a significant period of time is a rounding error.
- preinheimer 6y ago
  > Content moderation has been ~solved as well. I strongly disagree. Pornhub just purged 10 million videos because they couldn't moderate out child porn and rape videos. There's terrible content all over Facebook - Using Facebook's own numbers ~750,000 instances of "child nudity and sexual exploitation" were reported by users before being removed in Q3 2020. https://transparency.facebook.com/community-standards-enforcement#child-nudity-and-sexual-exploitation https://transparency.facebook.com/community-standards-enforc... - Facebook’s improved AI isn’t preventing harmful content from spreading https://venturebeat.com/2020/11/19/facebooks-improved-ai-isnt-preventing-harmful-content-from-spreading/ https://venturebeat.com/2020/11/19/facebooks-improved-ai-isn... - Despite Facebook ban on Holocaust denial, it’s still easy to find https://www.jpost.com/diaspora/despite-facebook-ban-on-holocaust-denial-its-still-easy-to-find-650292 https://www.jpost.com/diaspora/despite-facebook-ban-on-holoc...
  - ardy42 6y ago
    >> Content moderation has been ~solved as well. > I strongly disagree. Pornhub just purged 10 million videos because they couldn't moderate out child porn and rape videos. PornHub is not a good example here. Reactive content moderation doesn't work for porn, because it's too hard to distinguish a positive from a negative. Content moderation for porn has probably been solved...by using a per-clearance model coupled with strong identity verification. > There's terrible content all over Facebook That's probably true, but the worst appears to be much harder to find. It's a lot easier to detect and ban all porn (especially if you can ban all nudity) than just certain heinous kinds of porn specifically. > - Facebook’s improved AI isn’t preventing harmful content from spreading https://venturebeat.com/2020/11/19/facebooks-improved-ai-isn https://venturebeat.com/2020/11/19/facebooks-improved-ai-isn... > - Despite Facebook ban on Holocaust denial, it’s still easy to find https://www.jpost.com/diaspora/despite-facebook-ban-on-holoc https://www.jpost.com/diaspora/despite-facebook-ban-on-holoc... Facebook seems to have more problems with trickier things, like hate speech, holocaust denial, and mis/disinformation. We have pretty good technology now for recognizing things in images, but we don't nearly technology that good for understanding the meaning of speech and text.
AnthonyMouse 6y ago
> scale. This is the error. The problem isn't scale. If YouTube has 500 hours of video uploaded per minute and some other site has only one minute of video uploaded per minute, the smaller site is no better able to handle it, because they don't have proportionally any more staff than YouTube. The actual problem is marginal cost. If you have a Walmart store, it probably costs over a million dollars to construct it, and takes in several million dollars a year in revenue. A $10,000 fire suppression system may be expensive, but it's manageable. And if there are a thousand Walmart stores all over the country that each need a $10,000 fire suppression system, that scales just fine. You have a choice between an unsafe Walmart and a safe Walmart whose proprietors make slightly less profit. The marginal profit from hosting a given video is pennies. That means any solution that costs more than pennies isn't a choice between unsafe and safe video hosting, it's a choice between having video hosting and not having it. Scale has nothing to do with it; greater or lesser scale doesn't save you from a cost that exceeds your entire margin.

mytherin 6y ago

Fire suppression systems fail, cars get recalled or malfunction, people get food poisoning in restaurants or from food bought at a supermarket. Any product, not just tech, runs into problems when manufactured at scale. Tech is more visible because it is newer and in the spotlight. Nobody says "stop manufacturing cars" when one manufacturer issues a safety recall because cars have been around so long. The entire premise of producing 100X fewer cars so we can hand-craft cars and hand-check them is ridiculous - we simply accept the occasional manufacturing defect or product recall as the rates at which they happen are so small. These are all rounding errors, and the benefits of scale far outweigh the cons. The same applies to tech. Content moderation has been ~solved as well. The overwhelming majority of content on social media platforms, websites, etc is regular, legal content. The illegal content that makes it through the filters and stays up for a significant period of time is a rounding error.

preinheimer 6y ago
> Content moderation has been ~solved as well. I strongly disagree. Pornhub just purged 10 million videos because they couldn't moderate out child porn and rape videos. There's terrible content all over Facebook - Using Facebook's own numbers ~750,000 instances of "child nudity and sexual exploitation" were reported by users before being removed in Q3 2020. https://transparency.facebook.com/community-standards-enforcement#child-nudity-and-sexual-exploitation https://transparency.facebook.com/community-standards-enforc... - Facebook’s improved AI isn’t preventing harmful content from spreading https://venturebeat.com/2020/11/19/facebooks-improved-ai-isnt-preventing-harmful-content-from-spreading/ https://venturebeat.com/2020/11/19/facebooks-improved-ai-isn... - Despite Facebook ban on Holocaust denial, it’s still easy to find https://www.jpost.com/diaspora/despite-facebook-ban-on-holocaust-denial-its-still-easy-to-find-650292 https://www.jpost.com/diaspora/despite-facebook-ban-on-holoc...
- ardy42 6y ago
  >> Content moderation has been ~solved as well. > I strongly disagree. Pornhub just purged 10 million videos because they couldn't moderate out child porn and rape videos. PornHub is not a good example here. Reactive content moderation doesn't work for porn, because it's too hard to distinguish a positive from a negative. Content moderation for porn has probably been solved...by using a per-clearance model coupled with strong identity verification. > There's terrible content all over Facebook That's probably true, but the worst appears to be much harder to find. It's a lot easier to detect and ban all porn (especially if you can ban all nudity) than just certain heinous kinds of porn specifically. > - Facebook’s improved AI isn’t preventing harmful content from spreading https://venturebeat.com/2020/11/19/facebooks-improved-ai-isn https://venturebeat.com/2020/11/19/facebooks-improved-ai-isn... > - Despite Facebook ban on Holocaust denial, it’s still easy to find https://www.jpost.com/diaspora/despite-facebook-ban-on-holoc https://www.jpost.com/diaspora/despite-facebook-ban-on-holoc... Facebook seems to have more problems with trickier things, like hate speech, holocaust denial, and mis/disinformation. We have pretty good technology now for recognizing things in images, but we don't nearly technology that good for understanding the meaning of speech and text.

AnthonyMouse 6y ago

> scale. This is the error. The problem isn't scale. If YouTube has 500 hours of video uploaded per minute and some other site has only one minute of video uploaded per minute, the smaller site is no better able to handle it, because they don't have proportionally any more staff than YouTube. The actual problem is marginal cost. If you have a Walmart store, it probably costs over a million dollars to construct it, and takes in several million dollars a year in revenue. A $10,000 fire suppression system may be expensive, but it's manageable. And if there are a thousand Walmart stores all over the country that each need a $10,000 fire suppression system, that scales just fine. You have a choice between an unsafe Walmart and a safe Walmart whose proprietors make slightly less profit. The marginal profit from hosting a given video is pennies. That means any solution that costs more than pennies isn't a choice between unsafe and safe video hosting, it's a choice between having video hosting and not having it. Scale has nothing to do with it; greater or lesser scale doesn't save you from a cost that exceeds your entire margin.