undefined | Better HN

0 pointsrisyachka4y ago0 comments

Considering they are a data company with heavy AI departments - not be able to differentiate basic russian bots from users is very embarrassing.

0 comments

3 comments · 1 top-level

zacmps4y ago· 2 in thread

Most problems Google (& DeepMind) have seen success in are not arms races; you're simply trying to accomplish a difficult task.

Dealing with bots, and content moderation in general, is an arms race. You're fighting another person who could be just as smart as you in developing evasion strategies as you are in countering them. Getting it right all of the time is impossible.

dorgo4y ago

adversarial networks - isn't this exactly the branch of AI which is supposed to solve this?

zacmps4y ago

I'm not sure exactly what you mean by 'adversarial networks', it could be GANs or adversarial training.

Either way, yes, there are techniques for improving performance in the presence of an adversary but it doesn't negate the problem; anything you do will be countered eventually. Clever techniques only buy you more time before someone figures out a way around it.

j / k navigate · click thread line to collapse