Hacker News Reader

Comment by NotYourLawyer | Hacker News Reader

But “kill yourself” isn’t a keyword?

In my experience, the same string can result in different moderation based on unknown factors. For example, I used to report the trans slur of the "t" word and Facebook would remove the content. Now, when I report it, the comment stays posted. Maybe it has something to do with the age of the post or the number of positive reactions to it. Maybe I reported too much and the false positives resulted in my account being banned from reporting comments. Similar behaviors on other platforms.

> For example, I used to report the trans slur of the "t" word and Facebook would remove the content. Now, when I report it, the comment stays posted.

Isn't that also a slang term for a transmission? An automated rule for that that would probably result in far too many false positives: merriam-webster.com has the transmission sense as the first meaning, and has collected many very recent usages from auto magazines: https://www.merriam-webster.com/dictionary/tranny.

I couldn't actually think of what the 't' word was until you said transmission - that's the context where I hear it FAR more often.

To be clear, I reported it under photos of people who are trans while it was used as a slur in comments that were derogatory. Usually, this would result in the comment being deleted.

This is only an example to show there is no conspiration about moderation of gender related topics. You get the same issues with other slurs and other topics.

I suspect the majority of the "AI" for these moderation systems is "has this been reported by X number or Y% percentage of post viewers? then ban".

KYS has traditionally been a one way ticket to a twitter ban for sure.

With the impending Musk acquisition, perhaps the staff have lost their zeal and are preparing for more "free speech", as violent language and calls to violence online is often labeled.

Calls to violence aren’t free speech under any reasonable definition. At the same time, “violent language” feels like a contradiction in terms.

Well, if "Please dont kill yourself" triggered the "kill yourself" algorithm, that might be a problem too.

An automated moderation system should be smart enough to detect sympathy. And if it's in doubt, it should leave it to a person - a native speaker of the given language, because a lot of these systems are trained on English and western language and sentiments.

"Please don't kill yourself" can mean "kill yourself" depending of the original post and poster.

I know that my comment is a trivial observation, but an automated moderation system capable of detecting sympathy sounds like science fiction. Frankly it seems like humans struggle at it.

> An automated moderation system should be smart enough to detect sympathy.

That sounds like a super hard natural language understanding task.

> And if it's in doubt, it should leave it to a person - a native speaker of the given language, because a lot of these systems are trained on English and western language and sentiments.

That costs lots of money that could be going to the shareholders or executive compensation instead.

What if shareholders themselves are tasked to do that stuff?

yeah it is but "k1ll your$self" or whatever got through for a while, then it was "unalive", then "un@1ive" or sm when that got filtered, i've even seen "commit queen elizabeth".

trying to filter automatically based on message is a retarded strategy.

Reply on news.ycombinator.com