I’m training a neural network to generate recipes based on a database of about 30,000 examples, and although the network has managed to produce identifiable recipes, and even sometimes sort sweet from savory, it hasn’t actually managed to produce any good ones. Only a very few of