There are already many algorithms in the world that can generate a variety of works based on neural networks. They are still of little use in practice, and some are even used for fun. Last year, Janelle Shane collaborated with the New York Times to use AI to generate names for Halloween costumes. And recently, Will Knight of MIT decided to repeat the experiment, only in this case the same AI was used to create plots of Christmas stories. It turned out very ambiguous.
For work, it was decided to choose 360 Christmas films, the descriptions of the plots of which were taken from Wikipedia. The algorithm used in the work was the same as for generating the names of the Halloween costumes - textgenrnn.
The neural network has a lot of settings, but the researcher concentrated on three of them - "number of layers", "number of epochs" and "temperature". These are conventional names and they do not reflect what you think looking at them.
As a result, all the results looked extremely strange. Here is one of the synopsis. The translation is very approximate, because the neural network sometimes wrote words that were inconsistent with each other and did not follow the spelling and punctuation.
Sounds like gibberish and not even fun at all, unlike last year's experiment when the AI suggested some pretty funny names for Halloween costumes like "Minecraft Sexy Man." So the researcher also thought, starting to understand what the problem was.
Promotional video:
Although some sane and even ridiculous results were still achieved. Some of the synopsis turned out to be quite funny.
Synopsis
- The family of the Christmas terrorist moves to a new city and plans to celebrate Christmas.
- Ogre princess wants to cancel Christmas
- Santa Claus angel works in a factory
Some of the titles of non-existent films turned out to be pretty good and they even had posters for them.
Fight Christmas.
Christmas Mile.
It's Santa.
Vladimir Kuznetsov
