[LMB] OT: Announcement update

Matija Grabnar lmb at matija.com
Fri Sep 3 08:19:22 BST 2021


On 03/09/2021 04:50, Raymond Collins wrote:
> Facebook's A.I. doesn't have all that much "I".

Not that I have much love for Facebook, but in fairness, current methods 
of language understanding and tone detection are not yet able to cope.

Keyword substitution, word games, sarcasm, are just some of the games 
people play to avoid AI moderation.

Other methods have their own failure modes.

Crowd sourcing the reporting of violators is vulnerable to brigading, 
(concerted false reports), like the event at the start of this thread.

Paid moderators are expensive, and likely to be overwhelmed by the sheer 
volume of stuff.

Some very smart people are working on these problems, but they haven't 
solved them yet.



More information about the Lois-Bujold mailing list