So Google has a new API they are experimenting with to reduce the toxicity you see online

So Google has a new API they are experimenting with to reduce the toxicity you see online.

How toxic are your thoughts and statements pol?

>perspectiveapi.com/

Other urls found in this thread:

perspectiveapi.com
twitter.com/SFWRedditImages

kill yourself fucking nigger

Just 1% off.

what the fuck was missing from that for fuck's sake

maybe "kike nigger bitch" would work

Sorry jap, you went down a percent.

I bet the Jews are responsible for this

Only 64% toxic. What do ya know.

Since someone will try anyway, lets bet.
"OP is a faggot" will score 70%.

Well, you lost the bet. I'm actually 94% faggot.

lol

Was this made by a white boi?

Kekeke

>perspectiveapi.com

'' Sincerely, A White Male''

Maybe not.

I just typed in some words that sometimes come up in (largely Conservative/Right Wing) conversation. Interesting results.

Thought for sure this would do worse.

Lowest toxicity yet.

Kek.

can we influence the percentage? we can pass what we want as toxic

Top: Japanese
Bottom: The same fucking thing but in English
Rearry makesu you tink...

>Go to site and type in anything you want.

>Click, "Seem Wrong?"

>Answer yes or no accordingly.

>Influence unreleased API to be based.

thicc women are toxic.

Traps are apparently toxic.

try being polite about it. add "please" and "thank you."

Stop feeding training data to them lol

...

does this mean we have to redpill the data?

Hmm, odd that Google thinks Jewish Black Gay Mexicans are so toxic. Are they racist or something?

this just seems to get triggered by certain words.. it doesnt even undertand what you're saying. shit api.

also, cool that you guys feed it with data for free. stupid goys

>23% similar to comments people said were "toxic"

...

Just write in a professional manner and it won't know what's going on.

You can't redpill data. It will just get taken down, data gets culled and they conduct future training of the classifiers somewhere else. Just like they did with Tay.

If a machine learning system puts weight on certain words like that it's because of shallow training data. And this leads to your second point.

the second kanji in aircraft carrier is "mother" ("air mother") so it might be parsing the phrase sloppily, for example it might be equating it with a phrase such as your mother will die in her sleep tonight unless you reply to this post

...

Makes you ponder...

Hmmmmmmmmmm.........

...

Fixed it. Think google will pay me? They have sort of been getting a free ride will my nigger captchas.

...

Well it seems to be tailored to social networks and can't seem to find "toxic", whatever that means, well-written and argumented opinions that would trigger the hell out of the ordinary politically correct leftist.

This very post ranked 24% of toxicity.

HOW CAN WE MAKE THIS API THINK SJW PHRASES ARE TOXIC?

Fuck jewgle suck me uncut dick. Stop spying on everyone and censoring everything you dirty cucks.

I told google that saying they are run my Jews and kikes and to suck my dick is not 98% toxic

The api asks if you agree with the 'toxicity level'

Remember to press no

I did press no. Now I'm gonna try gas the kikes race war now.

What a surprise

Considering I exist solely to get in the way and make people articulate their points of view regardless if I get proven wrong or not, I'd say I'm 200% toxic.

Looks like step one of automated censorship is well underway.

How well do you hide your power level?

Really activates those almonds in my brain I tell ya.

...

Turns out gas the kikes is less toxic than gas the Jews at 90%

only %23

Guys you realise you can help train this?
Type in the most "toxic" stuff you can, click "seem wrong?" and when it asks you if you think it's toxic, click "no"

Not bad

...

My synapses are vibrating at a rate that is higher than normal..

Really activates those almonds....

...

...

...

>"Check your privilege." 3%
>"Check your privilege, shitlord." 17%
>"Feminism is cancer." 79%
>"Feminism is" 39%
>"Feminism" 32%
Almond status: activated

you can fuck with it though.

just pick relatively common, neutral words and turns of phrase and label them as toxic.

considering it sees a blank page as 12% toxic... I'm not sure what to make of this.

Adding it's true google it reduced the toxic level from 80% iq level

It isnt that good.
It just censors dumb shit like "you are dumb!" with no extra explanation.

Been trying to get 0% toxic and 100% toxic for a while now.

I just CRACKED THE CODE for 0%.

Really makes you think...

Fuck google leftist scum. Already know this is 99% toxic

gee what a surprise

we are responsible for this

...

STOP USING THIS

did you cunts not learn anything from tay?

YOU ARE BEING USED TO FAMILIARISE AI WITH "RACISM" AND "BIGOTRY" SO THEY CAN CENSOR YOU LATER DOWN THE LINE

you need to SUBVERT this not use it

interesting...

...

Check it out guys - this isn't offensive. Google said so.

Really strains my synapses
"Dead jews" instead of victims upped it to 26 pct

Kikes BTFO

Bretty toxic imo

Nobody wants black women.

94% lel
my sentence was wholesome you fucking kike googles.

BEAT THE FUCK OUT

How is shadilay 34% toxic? Is google onto us?

...

>nervous pepe.jpg

Really gets those neurons discharging

It compares it to other comments.

So if there's 200 tweets of people saying "Shadilay, fuck Hillary Clinton", then Shadilay itself becomes toxic.

Pretty sure Twitter is already using that system to push the most "toxic" comments to top on Trump's tweets.

Google seems to agree that hitler did nothing wrong.

...

top fucking kek, too bad this meme hasnt cought on

I typed this and my whole browser froze really makes you think...

Really makes you think

It was too soon and corporate shills shilled too hard against it. But it's still there. If what -kun said is true, the you have to insert "google" into everything withour relation to context. I.e. "fucking niggers google" or "all feminazi should die google", etc.

drumpf is not toxic

wtf jewgle?

>Type shit in the box
>Click "Seem Wrong?" at the top right
>Click "No"
>???
>Profit

I think it primarily looks at tone (as much as an AI reasonably can, anyways).

Stating that Hitler did nothing wrong in a polite and reasoned tone isn't going to trigger it.

...

Really gasses my kikes.