Janelle Shane’s neural network blog, AIweirdness.com, features algorithms that try to invent human things like recipes, paint colors, and Halloween costumes. Their struggles illustrate what artificial intelligence is good at - and where it has a lot to learn. She has only made a neural network-generated recipe once and discovered that horseradish brownies are about as terrible as you might imagine.
The Humor of Machine Learning

At AIweirdness.com I train machine learning algorithms to imitate human things - cookbook recipes, Halloween costumes, and knock-knock jokes. Their struggles produce a characteristic sort of humor and an unexpected creativity. We'll address questions like What is a reasonable project to tackle with a neural network? What alternative approaches might work better? What kinds of results might I expect from a given dataset? What do neural networks actually understand about the data they're given?