Neural Network storytelling

You may have heard about GPT2 and its generated Unicorn press release. So its been a while now since we want to make a post about text generation Neural Networks like GPT2.

Following a previous post on the subject, we believe that these could be the next Teatronika tool for life robot improvisation!  And if you are skeptical, take a look at what it can do in its online version talktotransformer.com. Just follow the link and input the beginning of a story. For instance, when we input the two sentences:

“Teatronika is a project which tries to bring robotics closer to arts. Teatronika is building performances in which robots are the only actors…” 

GPT2 generation continued (I repeat we didn’t write that, the text was automatically generated pressing a button. I mean really!):

So, we left it at:  “… are the only actors” and continued ”in the story, in a variety of formats, including on stage. Our program was an experiment in bringing people together to play with computers and to experience some of life as it actually happened by using the physical tools.

What is the idea behind Teatronika?

The idea is to bring robots and artificial intelligence closer to us. To find things that we need to know on a basic level (or, if you prefer, to know about us by understanding our behaviors) and also, how to work together and create a collaborative culture.

How have you been inspired by this idea in the past few years? What technologies or events have influenced the project your organization has been involved in lately?”

Mind-blowing…

So a bit of context GPT2 is a “trained large-scale unsupervised language model which generates coherent paragraphs of text, achieves state-of-the-art performance on many language modeling benchmarks, and performs rudimentary reading comprehension, machine translation, question answering, and summarisation—all without task-specific training.” (from OpenAI web). In this medium article there are more details.