Hacker News Reader

Comment by paisawalla | Hacker News Reader

We're not talking about the hardest content problems on the internet. Look at any of Elon Musk's (or anyone in crypto's) popular tweets and you'll see a flood of incredibly obvious bots that poorly mimic his account, spamming some shit coin or advertising a BTC address.

Also, you're really understating how good Tesla self-drive is. It's not full self-drive but it's a real achievement.

View on HN

Yes, and Youtube and Facebook both have similar problems. There is survivalship bias at play here, you do see the bots that manage to get past the spam protection, but never see the ones that don't. For all you know, Twitter could be blocking 99% of the spam. Of course, as long as there's money in it (and there is), spammers will constantly be coming up with new and unexpected ways to bypass your systems. It's a neverending battle, and if you think it's trivial, you are the one understating how hard the problem is.

I never claimed Tesla self-driving wasn't an achievement, but clearly they underestimated how hard that would be themselves, so I wouldn't be surprised if they'd do the same for content moderation.

There's a substantial qualitative difference between what gets through Youtube's and Facebook's filter, versus what gets past Twitter. They're not even practicing the current state of the art, as practiced by their peers.

> Youtube and Facebook both have similar problems. There is survivalship bias at play here, you do see the bots that manage to get past the spam protection, but never see the ones that don't. For all you know, Twitter could be blocking 99% of the spam.

If you spend 5 seconds on YouTube and Facebook and compare that to Twitter, you'll see a massive difference.

The spam bots on Twitter are incredibly unsophisticated. It's literally bots with the same profile picture and name as the person they're trying to spam.

On the other hand, YouTube spam bots are actually incredibly sophisticated. They're using GPT-3 or some language model to generate text and reply to each other. Like, sometimes I'll read a comment and not be sure if it was a spam bot or not.

Twitter is leagues behind w/ their spam filters.

Obviously YMMV but in 2021, every tech video I commented on would get these replies, which all had 18+ profile pictures. Clicking their profile every account had the exact same "playlist" with a porn site ad inside, and the channel header had the exact same website link. I kept getting these for months.

https://www.youtube.com/watch?v=1Cw-vODp-8Y

>Also, you're really understating how good Tesla self-drive is

How good is it?

It’s pretty good! Does the vast majority of my driving for me, except for roundabouts, gravel roads, and exceptionally snowy days (Scandinavia).

I do long road trips and feel less exhausted than I would otherwise.

Good enough often enough that people stop paying attention (and die when the car hits an obstacle the AI doesn't recognise)

Keep in mind Tesla self-driving users have a literal survivorship bias.

Not really. Deaths are too low to matter on a statistical level for “literal” survivor bias

It's so good that Tesla sued people for posting videos where it fails miserably. Really screams "confidence", doesn't it?

Did this actually happen?

I know there was one instance when an engineer using an internal build got fired for vlogging it (which is not at all unusual, Apple would also fire you for showing off an internal build).

Reply on news.ycombinator.com