solving global human problems with algorithms seems scalable 
pic.twitter.com/6vATTAXPM3
so i tried google's thing and anything in arabic seems to have a baseline 30% toxicity. these are direct translations. other languages?pic.twitter.com/UKGkiV0PHJ
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
seems like they trained only on English-language blogs where most use of other languages likely to be spam. And didn't message this well
-
yeah, not at all. the conflation of "natural language processing" with "english-language processing" is super frustrating.
End of conversation
New conversation -
-
-
do you have a link for me to try out?
- End of conversation
New conversation -
-
-
Seems to do it on non-roman alphabet languages. Tested "hello" in Greek, Russian, Albanian, Chinese, Japanese, and Persian, all gave 34%
-
It says in their blog that they hope to roll out to languages other than English. Must be confused by non-english characters for now
-
it would be great if the site itself mentioned that
-
Yep, maybe they'll update the page after the feedback. Google Blog link ishttps://blog.google/topics/machine-learning/when-computers-learn-swear-using-machine-learning-better-online-conversations/ …
End of conversation
New conversation -
-
-
similar for Japanese, but not Spanish so it's probably any non-Roman alphabetpic.twitter.com/ciXfCJaFlK
-
lmao never mind, Italian gets hit with the toxicity stick toopic.twitter.com/Jdoi7eZVor
-
gonna guess ~30-34% is about where any text that contains only out-of-corpus words pops out
-
- 1 more reply
New conversation -
-
-
BASELINES FOR THE BASELINE LORD! BINARY LOGISTIC REGRESSIONS FOR THE BINARY LOGISTIC REGRESSION BLOB
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Yes, typing in portuguese and Japanese gives me a 20-30% toxicity, regardless of whether the content is actually toxic or not.
-
After trying a bit with toxic and non toxic comments, the network is just giving random values.
End of conversation
New conversation -
-
-
"You are super cute" in Hindi is 36% toxic.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
same with Chinese
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
German apparently not as aggressive, though definitly more aggressive than english.pic.twitter.com/K9V8mn20zG
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.