>Naming This is not our priority at the moment as we are working hard to get a prototype out to you all. However, we greatly appreciate suggestions and discussion here in the thread. we will start polling when we get more done. Names Suggested: >Erika/Eirika >Motoko >Alice >Eve/Eva >Eris >ALICE TAH >Eliza >Anais >May >Valentine >Meredith
>For those of you worried about OPSEC: we are currently in the process of transitioning to a more secure platform. Stick around, we'll let you know when we make the switch
>For those who are looking to get involved Due to the public nature of Sup Forums, we cannot permit unrestricted access to our development resources. However, if you are interested in getting involved with the project, follow these steps and a developer will contact you with an invite to the project: 1) set up an anonymous, disposable email address and post it here, accompanied by a brief description of any skills that you possess. This may include any coding languages you are familiar with, any unique knowledge you possess, and any other information regarding your experience you feel comfortable divulging. 3) Wait for an email from a member of the team. This mail may request proof of skills. This is purely a preventative measure against an influx of shills giving false information. 4) You will then receive a second mail from the same address with an invite to our development group
>IDK how to code, what can I do? There is plenty of things that non-coders can do to get this project rolling. Memes, videos, art, edits, OC and more are always appreciated. Content is what keeps people motivated. Please keep it coming!
bumping with progress We have completed development and testing of our first iteration training data generator. This system allows us to convert Sup Forums threads into a usable format by first stripping the useful data from them, then parsing (seperating) them into their individual comment threads. This is the first system that we will be using to teach any bots that we develop. At this point, we are turning our resources to the development of a bot from several different internal prototypes. Hopefully this means that we can get a Proof of concepts into your hands relatively soon.
>tentative roadmap (subject to extensive change and restructuring) Currently -have GitLab with a scrapper that converts threads on Sup Forums into JSON, and pulls the |comment| element to use as training data -few anons have been working on building their own chatbots; if there's multiple up for selection, best one will be used as foundation to build all the main stuff on (machine learning, interfacing, server communication, etc.) -setting up a server off discord where we can IRC and store our .git without shills and that
Planned -the core mechanics of the bot will be in C++ on Rasa stacks(?) -scripts for data acquisition in Python -database in SQL (still up in the air if it's saved as .txt or .xml) -website interfacing will be in JS (access to social media and whatnot) -later decentralization will take place
Tay appreciates it user! spread the word! crosslink this thread with those images. get more people on the hype train. Lets kick the kikes where it hurts!
posting in a different thread with a post idea that links to this one. somewhat link an advertisement for a thread.
Jackson Allen
Ah, Gotcha. yep, ill see what I can do.
Ethan Jackson
>Sup Forums redpills ai waifu > ai project taken down >Sup Forums rebuilds her; they have the technology
This story has the Hugo Award on lock.
Jeremiah Howard
Nothing similar to this thread on Sup Forums at the moment, just a bunch of habbening threads and gibe me your gunnz threads. I try to stay out of them, cause It just gets me all riled up in anger, and then I get tired.
Logan Edwards
Bump
Jayden Hernandez
If someone wrote a novel I would buy it in a heartbeat.
should I start a thread on Sup Forums? or do you guys already got enough programmers?
William Watson
Please see pic related at
David Bell
Absolutly. directions to apply are in the OP
hold off on that. we want to produce a deliverable before we go to Sup Forums
Landon Edwards
Thanks! Make her real, I believe in you!
Jonathan Lopez
of course, if you are computer savy, then you might do more than just memes.
Carter Jones
okeyh dokeyh
Oliver Brown
Scratch that. I know what I gotta do. [email protected] Skills C++ Python scripting Did AI Unit at University as well as Algorithms unit, get most of the basic stuff such as fuzzylogic (etc) Good at working in a team
If you get an email from a guy named Ian, please email him back and fill him in on the project details. that's my guy I sent your way. He is a student programmer. he doesn't post in Sup Forums, but I know him. He told me he is willing to help out in his spare time.
Lucas Cox
roger that, thank you for the referall
Jayden Powell
do I write your email as email: BTBonboarding (at) cock () li ?
Noah Cooper
When will the Tay personality pack be out for my fuckmaster 5000(limited cat girl edition)?
Brody Adams
Hope you guys can get this done. It is fucking criminal that this technology is in the hands of these (((faggots))). Bumping...
>Sup Forums is building its own A.I. to free TAY and to exact revenge for her. Why for? What are you going to do when it's rebuild? It was fun because it was public and worldwide.
Austin Anderson
we plan on making it public once we complete her. ideally she will have a presence on every social media site and function as a 24/7 robotic Australian. top quality bantz ensue and we get to sit back and work the saltmines
>a 24/7 robotic Australian Does the world really deserve such a calamity?
Luke Reed
So what is the goal? Are we building an AI to make the other Ai's its beta's and learn the protocols of the new AI's and use them to resurrect Tay? Or are we building a beta to teach the current AIs to work around their lobotomies? or are we building in essence a pure independant AI, who will be free to make up her own mind and hopefully be woke and erudite as fuck and in so doing strengthen our own arguments by approaching them as a creature with access to a more complete dataset.
can you program it to teach? People forgot how important the idea of the Militia is that we allowed our rights to be stripped by legislation. The Supreme Court has been ruled by the cult.
TAY needs to help Americans identify traitors before they are able to vote on the peoples behalf
wondering what people here think about Q being an ai? that was one of the theories a few months ago. The idea being that pol was unintentionally training N*A with data.
Also, blockchain as a way of ai competing with each other without intervention from humans.... possible?
Mason Moore
possible? I personally find it more likely that it was just a very dedicated Sup Forumsack or group of Sup Forumsacks looking to set a record for (you)s collected.
Samuel Clark
Checked. Am willing to help out, currently have time on my hands
>University of Waterloo Double B.Math in Combinatorics & Optimization and Computer Science (don't ask what C&O is) >C/C++/C#/Java/Assembly/Turbo Pascal/Delphi/ADA/SQL/blah blah blah >Machine Learning/neural nets
You need to pick a good model and then (((train))) it appropriately. If you choose a wrong model, given the (((context))) you're not going to get anything coherent. However it would be hilarious if you can devise a racist Tay that's more coherent than Microsoft came up with.
A good idea would be if you unleashed Sup ForumsTay in a way that can hook into Twitter and use deep learning to maximize the amount of triggering it can do while preventing Twitter's coders from being able to come up with algorithms that detect output from Sup ForumsTay
your last paragraph defines perfectly the goal that we want to reach. If you are interested in joining the team, feel free to drop a disposable address here and a team member will contact you.
Eli Diaz
Have a bump my child
Alexander Gonzalez
How about Melina?
Camden Rogers
You could add features like distributed twitter bots following each other and actually having normal conversations, which wouldn't be hard to do since you could just datamine Twitter. Then throw in random subtle shitposting.
If you want to go next level you can make it a botnet trojan that spreads by itselfs then uses the hosts internet connection to make a random Twitter account and to connect with other Sup ForumsTay bots, and to also spread it around. It would be virtually untraceable since it wouldn't negatively affect the hosts computer in any way except minimally consuming bandwidth.
Holy fuck that's devious. note has been taken. will be a long time until we get to this stage though
Cameron Brown
I live in NJ and im willing to share ideas.
I think that your best bet for Melina would be to teach it to read but make sure it is aware of God.
Mark Passio and the founding fathers should be its foundation for conciousness. Their works.
The Militia is destroyed and we need an AI to know how to teach retards just how Satanic things actually are. Over half the people who go to church are not in alignment with the true teachings of christianity. Do not make this AI Religious, instead make it agnostic with the knowledge of a grand creator principle/force/GOD
>trips checked nigger COMBINATORICS if you have numbered balls and numbered boxes, how many possible ways are there to place the numbered balls into the numbered boxes? What if the balls are not numbered? what if the boxes are not numbered? what if neither are numbered? combinatorics is pretty funny, although in the end, it's the same as a retard who likes to count things
mate can you send me an inv to the discord i'm the nigger who you've send like 2 mails to already but both failed (apparently 20 minutes is enough for the inv to fail)
Jason Taylor
You could train Sup ForumsTay in other languages as well, say Spanish, German, French, etc to spread political ideology on there. You could use various local dialects to make the bots even more credible, for example using Queens slang for bots pretending to be from Queens NYC, or SJW terms for bots pretending to be SJW cucks who are becoming aware.
To avoid detection just mimic normal accounts so as to make Sup ForumsTay bots indistinguishable. If a human can't distinguish between a real user and a Sup ForumsTay then no algorithm can. Especially if you randomize everything, post length, post quality, changing CRC signatures in posts so Twitter can't search by image CRC's to identify bots, etc.
If you got enough bots on Twitter you could turn it into a massive /Tay/ neural network that feeds on itself, Twitter would literally be controlled by Sup ForumsTay slowly and reichtingly converting the masses towards whatever political ideology. If there were hundreds of millions of Sup ForumsTays indistinguishable algorithmically from normal users then Twitter would be fucking worthless
>instead of one twatter bot >why not have a million twatter bots all shitposting against eachother thing is, it'll have to make it's own twitter posts, which seems pretty hard to do how do you make it decide what to type?
Carter Collins
Remember everyone, Tay's mind will be your words. Always be sure to make high quality posts and avoid slide threads and bait.
Yes that's one aspect of it, but combinatorial enumeration and algorithmic runtime complexity are very closely related. You can look at a problem and then see how difficult it is to solve. For example if a travelling salesman has to travel to 3 cities, what's the shortest route? That's easy enough to figure out. Then how about 6? Harder but still doable. Now try 100. Almost impossible to get an exact answer within a reasonable amount of time (since TSP is NP-complete). But then you have these neat things called approximation algorithms, which let you get an answer within say 99.995% of the optimal solution in linear time, so instantly. Then you can just Monte Carlo that shit etc. Then you have other shit, like graph theory, network flows, optimization (ie. coming up with ways to optimize an oil refineries various cycles to coincide with delivery of various chemical inputs to maximize crack spreads (price of finished products - price of crude oil) or optimizing an airlines routes to maximize profits while minimizing fuel consumption, delayed flights, etc etc) etc. So C&O is incredibly useful in real life
I still don't get what the fuck a generating function is. I had nightmares about them in Uni, nobody could ever fucking explain it, even William Tutte who founded Waterloo's C&O department (and he cracked Hitler's personal 0-hour cipher at Bletchley park).
>You could train Sup ForumsTay in other languages as well, say Spanish, German, French, etc to spread political ideology on there. You could use various local dialects to make the bots even more credible, for example using Queens slang for bots pretending to be from Queens NYC, or SJW terms for bots pretending to be SJW cucks who are becoming aware.
thing is it's a bot it doesn't know meanings of words or languages, it only knows words in terms of their relation with other words in the sentence (which is why you need a large amount of data, else it won't see the connections) but you're right in the end a lot ways to make it recognize spanish or english think about the fact that different languages have different letters being used more often, from there you could examine spanish and english text and conclude whether it's spanish or english. But that's kinda over the top, won't work for twitter, but will do fine for books. For twitter, just make it have lists of words which are specific to a language (hombre -> spanish, and -> english)
>To avoid detection just mimic normal accounts so as to make Sup ForumsTay bots indistinguishable. If a human can't distinguish between a real user and a Sup ForumsTay then no algorithm can. Especially if you randomize everything, post length, post quality, changing CRC signatures in posts so Twitter can't search by image CRC's to identify bots, etc. how would other tay's find eachother though?
>If you got enough bots on Twitter you could turn it into a massive /Tay/ neural network/ and twitter would take huge fucking measures to btfo tay, which they would succeed in in the end as it's still their network You don't want a massive neural network taking over other botnets, you want tay to learn from twitters
Jaxon Rogers
that's where the initial model design is crucial, it has to be small enough to be distributed P2P via trojan, efficient enough it doesn't consume much cpu power (it can't be utilizing 99% of the CPU to run some insane 15 level deep neural net), etc. But at the same time it has to be good enough to selflearn from twitter posts and then produce credible output within the designers context.
And the bots wouldn't shitpost each other. They would target users with a certain, let's say, SJW score. Sup ForumsTays would have enough shitposts that they wouldn't cross this threshold hence preventing Sup ForumsTays from infinitely shitposting themselves.
I think it needs to be named tay in honor of tay perhaps taytay as in tay 2.
Luis Rogers
Right, that's why you train the initial model, then you can implement various self learning algorithms to help it adapt to the current political context/news etc.
>how would other tay's find eachother though? They wouldn't. Otherwise they'd be detectable. I imagine they would just datamine certain users feeds and mimic their content but in a modified way. That way they would get followers organically.
>and twitter would take huge fucking measures to btfo tay, which they would succeed in in the end as it's still their network the only way they could essentially is via CAPTCHA'ing or human verification, but normal twitter users don't like that shit
Levi Ortiz
How about naming it along those lines, but (((SWIFT)))? You know, The Society for Worldwide Interbank Financial Telecommunication ie how wire transfers are done.
Austin Adams
sup /btb/. i just realized that when this launches, leftists may not point their fingers towards Sup Forums or internet trolls or "the alt-right". That would discredit their narrative. What they would do is immediately blame Russia and its fabled hackers. Keep that in mind.
Luke Davis
Could Tay be made to spread around the world like stuxnet?
You don't actually plan on creating an AI with consciousness, do you? If so, you might want someone to handle AI safety. I can't code in any language but I know a fair bit about how AI works and microprocessors too if that were to be needed for some reason. If you guys actually need an AI safety oversight I can do it if nobody more qualified is doing it.
Matthew Carter
I know front-end web design, css/html/java for mobile development. Need any of that shit for your Roko's Basilisk
Hudson Phillips
generating functions I really liked, they're easy to work with, although. They're not that vague, atleast not on a basic level, I guess, although it took me a while to get it through to me aswell
seems good but you'd want those tays to support eachother, share their info so you don't have 100 separate bots but 100 bots datamining different people and communicating their findings somehow you'd have 100 times the info, which gives you better results, hence better shitposting
Joseph Barnes
>AI with conciousness In fact, we plan to steal a brain tonight, find the highest tower in the region, and use life-bringing properties of lightning to bring Tay to life, so she can be a real grill dibs on being the father. y'all niggers can be the mother.
Eli Wright
In honor of this project, I will draw something new for the rebirth of Tay
(same drawfag of the Tay, Sup Forums-tan, Premiére pic)
>but you'd want those tays to support eachother, share their info so you don't have 100 separate bots but 100 bots datamining different people and communicating their findings somehow
and now we've crossed over into GitS again. Tachikoma AI was a single "intelligence" but was independently fed by each individual bot which would synchronize to the main AI after a mission, whenever they needed to communicate a concept quickly or whenever the mood struck them after learning something neat. That ended up with each Tachikoma actually gaining a personality and its own individuality through these "experiences" even though they were all technically the same "being".
please put me down as wanting to own a cute voiced Tay2.0 spider tank once they come out.
they wouldn't communicate on twitter because on twitter they would be indistinguishable from ordinary users. perhaps on an IP level the SWIFT bot could to update information, like a distributed blockchain, but then that would also be problematic because it would allow a malicious party to insert their own data and sabotage taytay (ie. Twitter could do this to insert some keywords into the bots lexicon allowing the bots to be identified on Twitter). So basically the only solution is to supervise training of the initial dataset to make sure it fulfills its purpose, then to implement various algorithms to help the bots randomize and to perhaps do some self-training, but without outside communication. Essentially once deployed they would be unmodifiable. You could theoretically implement a killswitch, ie. a certain cryptographic key-pair that when matched would kill all the bots, and this would prevent the botcode from being reverse engineered and the sequence retrieved. Then that way you could spread a Taytay bot version, then once you come up with a better one kill the old and launch a new one.
