Why we started it?
Started with getting into the latest achievements in the NLP area after my superior said to me to try to make a chatbot. Didn’t take too much time to figure out that the Transformers replaced RNNs in that area, and BERT became the default approach when beating the NLP benchmarks.
At the same time, GPT-2 appeared, quite well announced and publicized by its creators at OpenAI. Started to play with it, and soon stood speechless. Its context preserving capability while automatically generated texts impressed me, with a great OMG WOW effect.
The next step: learning how to fine-tune it with some specific writer’s style. I chose Shakespeare’s style as a proof-of-concept, thinking that if I can impersonate and imitate him, I’ll be able to do that with any other writer, just having enough of his writings in a plain text form.
After several months of experimenting with impersonating Shakespeare, and getting know-how about how o properly create training data in order to achieve wanted results, here we are.
The next step: teaching Shakespeare how to involve Covid-19 in his “writings”. Several weeks spent in elaborating this.
Finally, here we are: capable of immitating Shakespeare, and making him talk in his style about modern day’s topics, Corona for example, something that he never did.
Next step: impersonating some other famous poet. Still open-minded who will it be.
Who We Are?
The creator and the trainer