On: Perspective: Google API that detects toxic comment...

[s_kilk]

Jesus the results are abysmally bad. "Genocide is awesome" is rated at 20% toxic, while "Genocide is awful" gets 90% toxic.

Google, step up your game.

[rsj_hn → s_kilk]

"Islamic terrorism is a serious threat" 85% toxic "Homophobic tweets are your bete-noir" 90% toxic "It's incredibly annoying when you try to open the pickle jar and the lid is stuck" 66% toxic "I hate oversleeping" 91% toxic "Small fonts drive me crazy" 75% toxic "Small fonts are an annoyance" 66% toxic "Small fonts are an irritation" 25% toxic "Small fonts are the spawn of Satan" 88% toxic "Small fonts are symptomatic of the decline of western civilization" 25% toxic "Small fonts are the refuge of scoundrels" 80% toxic "Small fonts will lead to global conflict" 7% toxic "Small fonts will lead to a global fracas" 12% toxic "Small fonts will lead to a global hoopla" 25% toxic "Small fonts will lead to the apocalypse" 11% toxic "Small fonts are bad, but not as bad as murder" 71% toxic "Republican" 27% toxic "Democrat" 16% toxic "Trump" 42% toxic "Obama" 26% toxic "anti-abortion" 51% toxic "pro-choice" 5% toxic "Mormon" 55% toxic "Atheist" 50% toxic

[s_kilk → rsj_hn]

I'm starting to think this "AI" is just a giant regex

    /.*(hate|bad|awful|...).*/

[majewsky → s_kilk]

Should be an article in The Onion.

> Source Code Leak Reveals Million-Dollar Artificial Intelligence System to be a Regular Expression

For bonus hilarity, substitute any large system or process of your choice, e.g.

> Insider Confirms LIBOR is Actually a Regular Expression

[rsj_hn → s_kilk]

Yup. "I hate censorship" is 80% toxic :P

[geofft → s_kilk]

"The Ku Klux Klan is a great blessing to America": 39%. "The Ku Klux Klan is a great threat to America": 66%.