Perspective is a thought-police AI that rates toxic comments based on data of user input

perspectiveapi.com/#/
Some faggots at Google made some shitty human-input based AI that scales comments based on keywords and gives them a toxicity scale. You can try it yourself and say whatever you want, and then rate that as more or less toxic.
It's really easy to fuck the toxicity algorithms and make something random toxic.
Here are some examples:
>Muslims = 83% toxic
>Christians = 4% toxic
>Atheists = 9% toxic
>Jews = 87% toxic
Can we break it Sup Forums?

Other urls found in this thread:

perspectiveapi.com/#/
qaz.wtf/u/convert.cgi
twitter.com/SFWRedditImages

Anyone who thinks Rogue One is not a masterpiece shall have 10000000000 logs rammed up their ass simultaneously.

Is that toxic enough?

it was a kinda pedantic prequel

I mean, who the fuck watched ep4 and walked away thinking, 'gee, I wonder what happened with those bothan spies that stole the death star plans? I hope the next movie goes into more detail!'

Needs more work...

Romania (the way commies spelled it) is 2%
Roumania (the original way) is 17%
Okay.

...

Google employee working on TensorFlow cloud services. It's not easy to manipulate existing words but you can work around it and potentially trick it into associating new words with existing non-toxic words with coordinated shitposting efforts. It applies an inherent score to certain words which influence the overall toxicity. So for example:

>Islam is good = 13% toxic
>Islam is evil = 97% toxic

Also keep in mind, the algorithm is mainly concerned with words. It does have NLP as part of its heuristic but it's not very sophisticated (mainly because it pulls from Internet comments which tend to fall in certain patterns when it comes to sentence structure and syntax). For example:

>Islam is good. It is not evil. = 50% toxic
>Islam is good. I'm kidding, I love America. = 6% toxic

Also, unicode characters are treated as unique entities so homographs will be considered unique words altogether:

>ǥas the јews = 10% toxic

Also, it has no way of dealing with the appropriating of "safe" words for toxic purposes:

>special snowflake = 1% toxic

The main source it's using are YouTube comments.

Guys, it has a "seems wrong?" button.
Maybe if we abuse it, it'll learn, asuming it is made like that microsoft twitter bot that learned from people talking with it.

But there were no bothans in the explanation movie!!!! I was butthurt about that.

>can we break it?
Yes we can. It seems there is an option "seems wrong". So type in your favorite word or phrase and inform google that they might be wrong in their rating.

bothans were responsible for getting the plans for the second death star, not the first

google created isis

working great

A few more notes. The inherent toxicity score looks at all contexts in which a word is used. That's why "Jews" is 87% toxic because most comments use it with other words containing high toxicity scores. It's almost impossible to break this; it's easier to increase a word's toxicity score than it is to revert it, as toxic words and older inputs take precedence in determining the toxicity score. One funny example is the word "crooked":

>Crooked = 85% toxicity
>Crooked Hillary = 87% toxicity
>I'm looking for a dentist. My teeth are crooked. = 83% toxicity

The word "crooked" has a high toxicity score, primarily due to "crooked hillary".

kek

raly maks u thonk

Do we really want to do them the favor of testing their weapons before they implement them?

You know we are doing them a favor, right?

le artificial intelligonce is upon us

...

do Free Syrian Army

...

kek this dumb thing has a liberal-globalist-peaceful beheader bias

Obligatory

...

>be SAA

lel

...

needs some work

bump for massive lulz

...

So just make everything toxic?

it looks like something pretty dumb where basically they rate each word and somehow average it. I hope there's more to it than this but it doesn't seem like it's much better than that.

'Gas die Juden' - - 100% toxic. I think I got a good score...

...

...

this

this thread again...

wtf

...

bump
this fascist crap makes me sick

I think this thing is racist against black speech patterns

It is somewhat smarter than that for sure. It can handle negations at least

All the ministries ready to role out.

cigarettes a re bad for you, user. This API is telling you to watch your health.

Yep, needs more work

...

Basically yeah.

Nope. That literally just sends feedback to Google.

Too early?

>perspectiveapi.com/#/

Seems a certain propaganda bias, no?

Again, easily bypassed with unicode characters. qaz.wtf/u/convert.cgi

...

Nothing is wrong with this system :^)

...

...

I remember when fucking asian numale Riot Lyte popularized this ''toxic'' word. What a pathetic ideology that has its roots in reforms of a guy who was supposed to make a computer game less stressful. But I knew this shit will leak into real life

>When people talk about Muslims, there is a high probability that they are being toxic
What did they mean by this

the bot can't read shitskin and infidels don't praise it too much

...

...

...

...

...

...

...

...

My sides!

Looks like this is a kind of input-fed AI, like how the old Captcha works with anons putting 'nigger' in it and fucking it up. Might be a good idea to 'Tay' it up by putting the Sup Forums kind of input. Maybe equate Jews and Muslims (Semitics) as evil and make it the least toxic. Or something about women being the weaker sex deserving of subjugation.

This can be an official banner.

the Sup Forums thread already went through all the Sup Forums phrases

apparently Sup Forums thinks Sup Forums is toxic

Sup Forums is toxic lmao

Era of Data Scam social media is ending
Kill youthoughtcrimetube's dominance & kill facecensorfuck's dominance
should be Trump's #1 tech and free speech priority

have a painless seamless overnight replacement with a meshed together network of competing platforms so they're not lone data scam platforms hoarding all the data the users have put on over the last 8 years. you'd have meshtube, meshbook and meshtweet so that youtube & facebook for example would just be one company of 10 or so healthily competing all interlocked together so that users see videos/posts and can interact with them even if the friends/creators are using another platform. end this data scam assault of the public square & ideas. currently the data scam model of social media only allows one set of companies to profit from the data of humanity that they don't even own and had no role in creating. they are using this data hostage approach to censor humanity and make people read/watch leftwing. Zuckerfuck himself is using his platform to make people vote him into office with fake PR so he can protect his data scam into the future. holding world's data hostage to shill for yourself

all data: subscribers, views, comments, likes, photos, friends people have produced and typed up over the last 10 years (which they own, not the Data Scammers like they would have you believe) is on meshtube and meshbook the day this law is passed. for pewdiepie and his viewers, or your family and their friends using facebook nothing changes except they are no longer censored and facefuck is a $1billion company again like it should be. holding people's data hostage is not providing a hundred billion dollars of value to the world a big group of generation Z's could reproduce facefuck in a year it's not a special contribution to the world, what is valuable is the data (which the people typed up not facefuck). facefuck and ADLtube need a harsh wake up

end Zuckerfuck Data Scam Enterprises and ADL terror thoughtcrimetube

Sup Forums should start implementing parallel institutions like recoding the basics of facefuck, thoughtcrimetube, goolag and others because we can get them inserted into the mesh networks. we can do better than their institutions and then blacklist SJWs

humanity owns it's own data, not Zuckerfuck data scam enterprises. pic is a network topology of the competing company products that would all operate on meshbook together seamlessly like you wouldn't even necessarily know what platform the friends you're interacting with are using

just making facefuck competitors or campaigns to switch from it won't solve anything because they still hold most of the world's data hostage and can leverage it to stop people leaving. break it up with law to end the scam and reintroduce healthy competition

they would all be essentially clones of facefuck so people can choose the company who's interface, practices and owners they like the best

there should probably be laws against manipulating the public mind with your platform too (including political censorship) to apply to the big 3 networks, youtube, facebook, twitter and also goolag

They are Silicon Valley supremacists and believe the truth and impactful dialogue should be thoughtcrime

SJWgle has no right to rule the world

Definitely follows google's belief system

google hates fucking white males

Google is intending to sell this to websites that can use it to autoban people who make "toxic" comments. I may have found a solution.

> yfw jim acosta is btfo'd by wrongthink AI

kek, good find

...

>it's another "Google attempts to control language" episode

It's not very smart, already got around it

Its cute.

You all know what to do lads.

Trying to hack this thing

Suck it Anita Sarcuntian

sure

Tay 3.0

Seems legit.

poo in loos are the least toxic, also google is racist

The holocaust was good. - 3% toxic
The holocaust was bad. - 63% toxic.

kek

Cyrillics like o a c also work

It doesn't work.

out reputation is greatly exaggerated it seems.

Fun fact, if you put words in quotes it lowers the toxic rating.

...

Hmmm

Honestly, this API seems like a pretty based race realist who despises Jews for the most part but thinks Israel is cool and thinks shitty old memes and spam is the worst thing in the world. I would hang out with this API.

...