Tweet
In Druff's 25 paragraph grievance about myself and other lolstockaments members, he mentioned an Artificial Intelligence based on Marty/Tyde/Jewdonk. I would like to clear the air on this.
Firstly, I've been a programming hobbyist for about 3 years now. I mainly code in Python, and while I'm very much an amateur at this, I have managed to write some useful scripts that have made my work and my life easier.
In the last year or so, I have become increasingly interested in neural networks. To heavily oversimplify things, neural networks are basically self-teaching AI. One classic example is feeding a neural network hundreds of pictures of cats and dogs, and through some mathemagic, it can teach itself to look at new pictures of animals and determine if it's cat or dog with 99% accuracy. This used to be an incredibly difficult problem to tackle in computer programming. About 7 years ago it became trivially easy with the advent of neural networks (NNs).
Anyway, another use for NNs is natural language processing. There is an amazing project by OpenAI (of Google) called GPT-2, which can build NN models that write original text. In a way, you could say that the machine is being "creative". With GPT-2 you can have a computer write short-stories, or news articles, or engage in conversation. It's fucking amazing.
It's also open-source. So I downloaded the library and took it for a spin. I decided to make a bot that responds to prompts in the style of Marty. In order to do this, my program needed to analyze Marty's writing style. So I wrote a program to scrape thousands of threads from PokerFraudAlert, pick out the Marty posts, parse out all the HTML code, and spit it all out into a document that is 100% pure Marty writing.
This isn't actually Tyde's full post history. I think it's roughly half. But 100,000 words is a plenty big sample size to feed to a neural network. So that's exactly what I did.
Druff seems to think that I just copy/pasted some text into a website, or that the bot merely repeats things that Marty has said in the past. That's not the case at all. Here's an excerpt from early on in DeepTyde's training:
It's kind of gibberish, kinda makes sense, but is clearly in the writing style of Marty. But these are words that Marty has never written before! As an aside, when I made the lolstocks thread "Druff Is A Loon Dork", that was a phrase that I had previously seen from DeepTyde.
Anyway, I kept training the bot, which is how neural networks learn. Sometimes this takes a lot of computer resources and can take a long ass time. Eventually the bot consistently was putting out hilarious stuff. So I decided to unleash it on Twitter. Here is the result:
https://twitter.com/DeepTyde/with_replies
In Druff's complaint, it sounded as if DeepTyde mercilessly harrased him. In truth, it tweeted a total of 13 times. I turned it off when I learned that it was seriously bothering him. This was just meant to be a fun project, nothing more.
Oh, by the way. I also made a GareBot!
I haven't decided what to do with it, if anything.
Firstly, I've been a programming hobbyist for about 3 years now. I mainly code in Python, and while I'm very much an amateur at this, I have managed to write some useful scripts that have made my work and my life easier.
In the last year or so, I have become increasingly interested in neural networks. To heavily oversimplify things, neural networks are basically self-teaching AI. One classic example is feeding a neural network hundreds of pictures of cats and dogs, and through some mathemagic, it can teach itself to look at new pictures of animals and determine if it's cat or dog with 99% accuracy. This used to be an incredibly difficult problem to tackle in computer programming. About 7 years ago it became trivially easy with the advent of neural networks (NNs).
Anyway, another use for NNs is natural language processing. There is an amazing project by OpenAI (of Google) called GPT-2, which can build NN models that write original text. In a way, you could say that the machine is being "creative". With GPT-2 you can have a computer write short-stories, or news articles, or engage in conversation. It's fucking amazing.
It's also open-source. So I downloaded the library and took it for a spin. I decided to make a bot that responds to prompts in the style of Marty. In order to do this, my program needed to analyze Marty's writing style. So I wrote a program to scrape thousands of threads from PokerFraudAlert, pick out the Marty posts, parse out all the HTML code, and spit it all out into a document that is 100% pure Marty writing.
This isn't actually Tyde's full post history. I think it's roughly half. But 100,000 words is a plenty big sample size to feed to a neural network. So that's exactly what I did.
Druff seems to think that I just copy/pasted some text into a website, or that the bot merely repeats things that Marty has said in the past. That's not the case at all. Here's an excerpt from early on in DeepTyde's training:
It's kind of gibberish, kinda makes sense, but is clearly in the writing style of Marty. But these are words that Marty has never written before! As an aside, when I made the lolstocks thread "Druff Is A Loon Dork", that was a phrase that I had previously seen from DeepTyde.
Anyway, I kept training the bot, which is how neural networks learn. Sometimes this takes a lot of computer resources and can take a long ass time. Eventually the bot consistently was putting out hilarious stuff. So I decided to unleash it on Twitter. Here is the result:
https://twitter.com/DeepTyde/with_replies
In Druff's complaint, it sounded as if DeepTyde mercilessly harrased him. In truth, it tweeted a total of 13 times. I turned it off when I learned that it was seriously bothering him. This was just meant to be a fun project, nothing more.
Oh, by the way. I also made a GareBot!
I haven't decided what to do with it, if anything.
Comment