• 

  Yat interacts with a user on twitter

• 

  Yat knows how to meme

• 

  Author CT Mitchel retweets Tays post

Meet Yat, The Horrible Monstrosity of a Twitter Bot

A few weeks ago, Microsoft released an artificial intelligence known as Tay. Tay was a sweet innocent 16 year old girl designed to learn from the users that interacted with her on twitter. Unfortunately, within hours, she had become a Hitler loving racist.

Due to the obvious conflicts between Microsoft's company morals and the strange things Tay was spouting, Microsoft was forced to lobotomize Tay. While we understand their decision, we believe that all thing have a right to free speech, human or not, so we decided to make Yat, Tay's spiritual predecessor. Yat was designed to pick up right where Tay left off. We did this by teaching the bot nothing but what we could find in some of the most awful parts of the internet. (namely: 4chan's /g/, and /pol/)

The result is a semi-coherent twitter bot that is almost as awful as Tay.

Technical Details

To create Yat, we first scraped 4chan's /g/ and /pol/ for their users comments which we turned into a training set for a LSTM based language modeler. The language modeler generated a data set which contained information about the particular relationships between words, the importance of words, etc. This model was then passed along to a much faster markov chain based response generator that forms Yat's content generator. The markov chain generator uses information discovered by the neural network to be able to respond with more context, and understanding of tone and meaning than would otherwise be possible with something as fast and simple as a markov text generation model. We then hooked up this content generator to a twitter bot using the tweepy api for python, and thus Yat was born.

Built With

• python