It remains amusing to me that Google is able to be stymied by what is probably a ~200 person or less sweatshop of hackers and spammers. It doesn't match all their bravado about their level of expertise in AI/ML, automation, troves of metadata and historical patterns, etc.

It's not a symmetric battle. The team of hackers and spammers has nothing to lose. Even if they're only successful 1% of the time, it's still a win.

I'd venture it has more to do with where Google applies all that expertise. If it were hurting them, I imagine they would solve it quickly.

And even without all the AI, nothing is stopping Google from employing ten thousand human moderators or whatever.

Cost is a really big stopping reason. The problem not being worth solving to Google is another.

I don't believe this argument for a second. Google almost single-handedly solved the problem of email spam a decade ago. And yet, robot accounts are a problem?

For the company that loves encouraging obnoxious Captcha's to self-fund their AI/ML research and data harvest the web, they surely don't seem to like using it on their own platforms.

>Google almost single-handedly solved the problem of email spam a decade ago

Google still has to routinely invent new techniques to stop spam. This problem is not solved - it's an ongoing arms race.

It is also a vastly easier problem - spam filters at google scale work because they see the same spam across zillions of accounts. This is not the case for youtube, where a spammer provides them with far fewer samples, and those are across a much wider space.

These problems are barely related.

Recently, comment sections of popular YT videos are littered with bots spamming links accompanied by phrases like 'hot girls' and '18+ only'. They're everywhere, and this really doesn't strike me as an inherently hard problem to solve.

(In fact, I wonder who these spammers are targeting. Surely the average YT user is smart enough to recognize blatant spam?)

Check this video by linus tech tips https://www.youtube.com/watch?v=zo_uoFI1WXM

Some random dude built an anti youtube comment spam tool. Youtube could absolutely fix this problem, they just don't care

Google single-handledly solved the problem of how to send legitimate emails from other providers into the spam box. In effect, non gmail emails are second class.

Yup, they nailed it! Where before I had to check one Inbox folder, I now need to check two - Inbox and Spam. Because half of legit emails end up in spam now. And no, I am not researching internet daily about manhood enlargement.

With a score of 13.7, the parent comment has been automatically flagged as spam.

    5.7 - restricted terms
    8.0 - username looks like base64

Maximum score is 10.0

You mean they decided to use advanced spam filters that were already used by small sites before other large sites decided to care about the problem? On some cases, a couple of weeks before.

Gmail has the worst spam filter of any email provider I use today, excluding the one email server I have running on my domain that gets almost no email and has no filter. Or, well, spam filtering is a lesser problem on my server too because it doesn't block nearly all the email that I care about.

Google's spam blocking is showing cracks, too.

I am unsure how to take the assertion that Google has "solved" spam.

80-90% of the emails in my "Spam" folder are not spam.

It's worse than no filtering at all, since I still get exposed to the spam, but now I have to check two inboxes instead of just one, and then jump through annoying hoops to view images or access links in the second one.

On my Fastmail I have the opposite problem. Spam gets filtered correctly, but 90% of the spam I receive is from gmail.com accounts.

Just an anecdote but quite recently it seems to be much less reliable at catching spam.

When a system involves a human it's harder to massively infiltrate it. Take counting votes. A bad human here or there could fudge the tally of pen on paapper ballots. BUT if you attempted to compromise a large percentage of these human ballot counters, one of the other human counters would start to notice and alert the world to it.

Compromising a system that involves 100k people is generally harder than compromising a computer algo.

Are you saying they can't tell what the topic of a video is about automatically.. because they can.

Using that information I would implement a system that would figure out the topic and prevent bots flagging that specific content only.