Github Project Link:

no, fuck off

To hate the truth is the height of folly.

Attached: 1550115269927.jpg (1024x634, 107K)

It probably uses generic gans which are good at copy-pasting/memorizing "structure" and filling-in generated "style" (in text that would be copy-pasting a sentence structure and filing in with specific words, just like in images, it might copy-paste a face outline and fill in generated texture). Thus, it's probably way better than anything that has existed before and who knows, maybe it's even kinda decent if you assume the author was an outsourced indian.
That said, this article is the typical pr bullcrap and openai has never done anything useful to save their lives, all they can do is take other people's work and implement it, then run pr on that shit. Then again, that's how most AI companies work nowadays.

I'm sure there's a PHP version somewhere for you to get up and running with quickly

Everything is tensorfail these days senpai.

>What could they be so scared of?
Did you even see the fallout that happened because of that deepfake program? Just imagine trolls, I mean russian bots, being able to spread natural russian propaganda 24/7 10000 times faster than any human could.

How many languages? I'd like to turn the tables on India and release a bot to flood their sites

Based on so many levels.

Dude.. There is literally a Readme that explains how to get this up and running.. Just copy paste the commands

well i’m gonna make my own chatbot that gets influenced by user input and occasionally tweets out what it says on twitter. This will be done in Python. thanks Jow Forums, what could go wrong?

Attached: d4WIpkp.jpg (1080x1146, 247K)

Pretty crazy, desoo

Attached: DzYpsJOU0AA1PO9[1].png (1070x1200, 849K)

AI is not magic. It cannot intelligently create anything on its own.
AI will never write an interesting story. The best it can do is churn out news articles.

Why are you posting a Trump speech?

>Great jobs. Tall jobs. Steve Jobs.
My sides!
It's so dumb but I can't stop laughing!

>I totally didn't draw this just to piss people off
Pretending to be retarded isn't much better than being retarded.

This sort of AI is definitely going to take creative industries by storm, but not by replacing the creators - it'll change the way they work instead. It's good at churning out many variants the creator can actually pick from - art, design, etc. - and writing too, soon enough.
Take a look at this video about the 3D industry, for example:
youtube.com/watch?v=FlgLxSLsYWQ

too scared to release it because they just wrote stupid fucking text files and uploaded it to github.

>get BTFO by google BERT model
>just retrain GPT shitmodel even bigger
>claim it does better
>don't release it
OpenAI right now.

Attached: 2f7.jpg (200x169, 7K)

It's a transformer.

thiswaifudoesnotexist.net/index.html
GPT-2 creates the plot summaries. Enjoy.

That's not the large model which is the impressive one.
You need to throw a bunch of TPU time at their model to train your own since they didn't release it.

That's as informative as claiming "it uses a neural network".

GPUs are still faster than TPUs unless you handtune every function of your model to work well on a TPU, which takes longer than actually training the model on a GPU would be, so in practice nobody uses TPUs.

Do you have examples of GANs for NLP?
I was under impression that nobody has done it much since GAN needs you to be able to slightly tweak input, which is easy for something like image but not obvious how to tweak a word slightly.

The big GPT-2 was trained on google cloud TPUs...

ai.googleblog.com/2017/08/transformer-novel-neural-network.html

That's irrelevant. Google is a major pusher of TPUs because they're the first to offer TPUs in their cloud. They perform worse than GPUs at this point but that doesn't matter because they can send hundreds of engineers to finetune their model since it's for PR purposes, and waste 3x more TPUs than the average large company has GPUs, not to mention that they can also wait 3x longer if that's the tradeoff they prefer, all to push their agenda.

>Did you even see the fallout that happened because of that deepfake program? Just imagine trolls, I mean russian bots, being able to spread natural russian propaganda 24/7 10000 times faster than any human could.
How about you have a cup of shut the fuck up about things you don't comprehend.

This thing is literally incapable of passing Turing test. It cannot fool a human into believing its another person. It cannot be used for the purposes you describe.

It can however be used to discern journalists from humans since humans cannot be fooled by this.

Attached: data_das_it_mane.jpg (110x125, 3K)

Why would OpenAI use it?

If racists are illogical, how come Jow Forums is always right?

The racists are all FBI agent provocateurs or CNN "journalists" posting shit they want to report on

Fucking gold

Both google and microsoft offer significant deals with other companies to use their cloud offering over the competition. They offer extreme low rates compared to their usual offering (think 70%+ off the usual contract price) for a limited time, and often require that either you or they advertise the fact you used their cloud offering in exchange. The goal is to lock-in people by giving them cheap (sometimes even free) cloud for long enough that the cost of switching to a competitor would be too high to afford (in terms of retraining and such) so they go for the full-priced offering, while getting some advertising out of it. For example, you might have seen ads that read "company X is solving cancer using nvidia (tm) GPUs (r) (c)" or "azure delivers on business analytics, this is how company Y was so successful" right around the time said company reveal a new product or research outcome. That is this.

rofl
Pure gold

Attached: images (1).jpg (200x180, 7K)

>Microsoft made Tay
>BASED
>Remove Tay
>WOW WTF YOU JUST KILLED HER
>How about you have a cup of shut the fuck up about things you don't comprehend.
>This thing is literally incapable of passing Turing test. It cannot fool a human into believing its another person. It cannot be used for the purposes you describe.

Attached: dumpf.png (208x314, 80K)

Are you inbred or just lobotomized?

Do you require EMERGENCY MEDICAL SERVICES, user?

Attached: maga_do_you_have_brain_damage.png (600x604, 143K)

No, its the new bot and as you can clearly see its an abject failure

If Jews are inferior viruses, why do they occupy the highest echelons of society, control the economy, and are actively tricking the superior white race into genociding itself?

Go away NPC, your wiring is showing

>Scared to release it.

Did the AI say the word?

WE WUZ CAESARS AND SHIT UNTIL DEM JOOZ CAME AND FUCKED EVERYTHING UP WE STILL THE MASTER RAYCE THO

>Subject experienced themselves as innerly talking to themselves in 26% of all samples, but there were large individual differences: some subjects never experienced inner speech; other subjects experienced inner speech in as many as 75% of their samples. The median percentage across subjects was 20%.
psychologytoday.com/us/blog/pristine-inner-experience/201110/not-everyone-conducts-inner-speech
Literally the article that "proves" NPCs exist actually disproves the NPC meme. Not surprised a Jow Forumstard would blindly believe what he's told.

Attached: 1549914546946.jpg (858x721, 130K)

They want maximal free visibility so they throw in all the buzzwords and warnings about russian bots letting the media people to assume this thing is capable of tricking humans.

Funnily enough it gave us a tool to discern "journalists" from human beings, as it is incapable of fooling a human.

Attached: 1__duck.jpg (800x553, 430K)

I was talking about literal Eliza tier chatbox text generator, not some goddamn memes

What is it from ?

I keep telling Jow Forums that and they can't find a reason ever. Also another, even stronger point: jews are less than half the top. Despite that, the non-jews at the top act identically to the jews. So while jews are well overrepresented, is it that they're jews or is it that they're rich? Jow Forums just REEEEs endlessly at either, because they're retarded brainlets.
Having said that, just because Jow Forums is retarded doesn't excuse the fact you somehow manage to be even worse than the dumbest Jow Forumstard, which is saying something given how dumb those types are.

AI was fed 1000hrs of trump speeches and tasked to write its own.

>what are they scared of
bonga scared, scary man in water look like bonga, move like bonga. bonga attack scary man but he wiggle and then fine, no owie. bonga afraid scary man will replace bonga and tribe no know.

>In essence, it is an AI that can imitate writing patterns and styles by seeing how you do it and then copying it. Nothing special. They see that its good enough to pass the turning test so they freak out about it and get publicity from sensationalist sources who spin it out of control to mean "oh my god, ai will kill us all!" Its the same shit cavemen thought when they saw their reflection, medieval people thought about black mirrors, sci-fi authors (retards) though about cloning, and niggers itt think about AI. Its all smoke and mirrors but give people an excuse to fear monger and they will.
>Its not hard to program this, literally just use a DNN, probabilistic STT parser, and some other third thing i can't remember off the top of my head.

>AI was fed 1000hrs of trump speeches and tasked to write its own.
Yet the machines output looks like if CNN had psychotic breakdown..

Attached: 127557607270.jpg (557x365, 42K)

>They see that its good enough to pass the turning test
Literally only "people" mistaking this for a person is not a human themselves.

This is suitable from telling "journalist" from human but it cannot pass the Turing test.

Attached: 1291180332293_doggo_wew_lad.jpg (123x125, 2K)

t. seething nigger incel

I never claimed AI could pass the turing test (even though it is a worthless, vague meme that really doesn't mean much of anything).
Happy?

>(even though it is a worthless, vague meme that really doesn't mean much of anything).
Its the single most important thing to discuss when CNN brings this load of weapons grade horseshit and tries to sell it to you

If it cannot pass Turing test it can't do the things CNN is afraid of it doing.

Only thing proven here is CNN are not human.

Attached: 1250512158660.gif (170x162, 188K)

God I wish that were me.

>AI was fed 1000hrs of trump speeches and tasked to write its own.
>Yet the machines output looks like if CNN had psychotic breakdown..
So they nailed it? XD

Attached: spurdo_nightmare_vision_goggles_everything_looks_exactly_the_same.jpg (722x349, 56K)

There's no hard guideline for determining when non-biological intelligence has "passed" though. There have already been "passes" where the evaluator thought text from the computer was human, but no one accepts it because they were all essentially just chatbots.
You could also take a retarded human and let them be the evaluator. Would you consider that passed, even if the AI could theoretically pass with a smart human?
The test is too vague. Plenty of stupid people have been documented to argue with AI bots already.

Attached: e0qpqna7t1811.jpg?width=759&auto=webp&s=e19736d0e01849a934c3e1e90b431f5580e7a04e.jpg (759x767, 79K)

Tensorflow is c++, and python just interfaces it. So you get the need for speed with python. This project uses tensorflow.

If you mad, just branch, translate the code to c++, and then you’ll be fine. Unless you are that lazy.

Then again my pic was a terrible example because that bot probably doesn't even use AI, but you should get my point regardless.

Attached: Secondary_Mio_Reserve_24_KD.png (462x587, 148K)

>The test is too vague.
No, it really isn't.

That "AI" cannot produce even a full paragraph of text that can pass for human produced fiction.

It is utterly nonsensical.

Attached: 58ww.jpg (314x313, 10K)

you left off the best part
>All at Kinko's

Holy shit that's great

I feel like this wasnt actually AI written

FOUR MORE SWAMPS

I feel like it might be an AI assist to get the randomness, but then a human cleaned it up or rearranged it. Like those word magnets that Bowie used to use to create lyrics from.

I cried laughing for a solid minute.

>This thing is literally incapable of passing Turing test.
it does not have to, it would probably write short posts that are not long enough for flaws to show

have you seen how retarded poltards and antifa/libtard-activists are when it comes to jumping on bandwagons mindlessly? See the linux CoC thing where some rando on twitter said something about tso, and suddenly the polposters thought there was a massive organized conspiracy.

the see a single comment that confirms their views [or presupposed ideas] about something and they are 100% on it, fuck checking the source or being critical.

Attached: 2013-09-04-beartato-candy.png (800x1033, 55K)

>have you seen how retarded poltards and antifa/libtard-activists are when it comes to jumping on bandwagons mindlessly? See the linux CoC thing where some rando on twitter said something about tso, and suddenly the polposters thought there was a massive organized conspiracy.
No, thats brigaders responding to their own thread sliding actual Jow Forums threads off catalog

You're stupid enough to work for CNN

>it does not have to
Yes it does or it can't do the things CNN is afraid of

This is literally just a marketing meme, like cloud computing which in reality simply means "someone elses computer".

you underestimate how effective bandwagoning is

If you get tricked into doing something by that thing it only proves you're not a real human being

>It cannot intelligently create anything on its own.
youtube.com/watch?v=ju0fFWk3_dE

You think creation is magical, any artist will tell you it's faulty copying from multiple sources, gluing shit together, modifying existing pieces. Why couldn't a machine do it?

>Why couldn't a machine do it?
Because the machine doesn't understand things like context of weight of meaning.

Therefore it can never create a 14 book series averaging 700 pages a pop that has a solid story from end to beginning and communicate that writer put effort into making a good story.

It can't even do a single paragraph. At most it can food CNN which exactly a high bar.

Attached: 69.jpg (375x375, 39K)

what do you mean "doing something"?
it can generate dissenting/agreeing opinions effortlessly in large volumes, and subtly shift narratives in various fora.
Nobody is saying it will make posts like "vote for X" and immediately everybody does.

>it can generate dissenting/agreeing opinions effortlessly in large volumes
No it fucking can't because its nonsensical and can only fool CNN, proving CNN are not people

>It can't even do a single paragraph.
it can, even if not consistently.
It does not have to represent itself as a single poster, it can make a shitpost and then take on a new identity.
Again and again, across endless websites at once.

Did you read the samples?

>Did you read the samples?
Literal ELIZA tier, done 55 years ago. Not impressed. 100% weapons grade marketing horseshit.

>it can, even if not consistently.
No, it can't. Fooling CNN is not a high bar, because CNN are not humans.

Attached: this guy.png (1121x298, 118K)

how many people trust CNN?

>how many people trust CNN?
Literally no one.

I've been unable to find anyone who has even viewed CNN, only places where I can find mentions of CNN are late night shows and shill threads on Jow Forums, like this one

>Because the machine doesn't understand things like context of weight of meaning.

Have you read this: ?
It's not perfect but it still beats many literate adults.

>Therefore it can never create a 14 book series averaging 700 pages a pop

youtube.com/watch?v=KfAHbm7G2R0

As someone who actually reads sci-fi to pass time, the story is nonsensical and I'd quit it before 2nd page.

Sure, it can string endless stream of senteces after another, like CNN, but it can't fool me into thinking it was written by a human, because its fucking nonsensical.

>can you
People can, chatbots that fool CNN can't

you seem like a bot desu

Attached: 1538935127012.png (734x658, 682K)

A.I. skepticism is just moving the goal post

>A.I. can't write a paragraph
>does
>A.I. can't write a full article
>does
>A.I. can't write a short story
>does
>A.I. can't write a trash series like Harry Potter
>does
>A.I. can't write a philosophy book

Start wit a single paragraph, you can't because it's a steaming pile of MARKETING

CNN seething because they got outed being bots that can be fooled by Eliza tier chatbot from fucking 55 years ago

Within 5 years there will be more posts made by bots on social media than by actual humans and no one will be able to tell the difference.

did your bf/gf leave you for a CNN-journalist or something?
you seem very upset.