Hacker News new | ask | show | jobs
by _fat_santa 1543 days ago
I think you're still going to bump up against scaling issues with this solution. Youtube has something like 500hrs of new content uploaded to it every minute, and I've heard speculation that this solution couldn't possibly be solved by humans because there simply isn't enough of us on earth.

In a way I sympathize with Google, Youtube, Facebook and all the other companies operating as mass scale. On the one hand their algorithms suck for moderation, but on the other hand they are the most advanced created yet and there is simply no alternative. I would imagine in the next 10-15 years the players will get moderation under control and be able to model AI to moderate fairly. But that's going to be a slow process because there isn't any alternative to AI in this case, at least how I see it.

5 comments

> I think you're still going to bump up against scaling issues with this solution. Youtube has something like 500hrs of new content uploaded to it every minute, and I've heard speculation that this solution couldn't possibly be solved by humans because there simply isn't enough of us on earth.

That's not the problem that needs solving. A manual review of the videos that are flagged by the AI snitch is all that's needed. Google can very easily afford this level of manual oversight.

> I think you're still going to bump up against scaling issues with this solution. Youtube has something like 500hrs of new content uploaded to it every minute, and I've heard speculation that this solution couldn't possibly be solved by humans because there simply isn't enough of us on earth.

Doesn't his means they've scaled beyond sustainable? As I read it, it's a good argument for them being broken up.

Breaking them up won't reduce the new videos total, or increase the number of humans on earth. So Breaking them up doesn't help.

I think a 'bond' system where you put up money to get human appeal, and get half back if the appeal is successful. Maybe with increasing bond as you lose more appeals. Is a much better system.

> In a way I sympathize with Google, Youtube, Facebook and all the other companies operating as mass scale. On the one hand their algorithms suck for moderation, but on the other hand they are the most advanced created yet and there is simply no alternative.

I don't. There's another alternative.

If you're growing too big to moderate, stop growing. Turn off signups. It's irresponsible to do anything else but, and "but we'll make less money" isn't a valid argument to this.

Why not charge for human review and offer a guaranteed refund if it turns out the user wasn't at fault.
Because then you're putting a price of entry on having basic support even when you fucked up. Any user from a country where even 5 bucks is a big deal is basically locked out of it. Any user from a country where Google doesn't take payment from is locked out of it.

When Google provides dogshit services, you expect people to pay for them to be reminded to fix it ?

They don’t have to charge the same price everywhere as the work doesn’t cost the same everywhere either.

Having to refund people in cases where Google’s AI fails would create an incentive for them to improve the AI and set higher thresholds for content to get flagged.

I think my idea would make the system work better overall without making it worse for anyone.

>Having to refund people in cases where Google’s AI fails would create an incentive for them to improve the AI and set higher thresholds for content to get flagged.

Here's how it would actually go:

* AI flags your video

* You pay 5 dollars to have it reviewed by a human

* Either immediately as it's conceived, or a few months down later the line when some manager figured out they could lower costs this way, an AI reviews your video again, and never finds Google at fault.

Never underestimate Google's absolute disdain for its users.

How are there enough people on earth to create all that uploaded content? Even cutting out and uploading a quick clip from a show or something takes time.