I’m training a neural network to generate recipes based on a database of about 30,000 examples, using the open-source char-rnn framework.

People have asked me whether I’ve tried making any of these recipes.

The answer is no. No, I most definitely haven’t. I don’t know, in fact, of anyone who has dared.

The following selections of ingredients and directions may help explain why.

Ingredients/directions not recommended:

10 lb dried broccoli casserole
1 lb crab water
¼ oz can of fish casings

Place wrapper in microwave on Low Coals.
Mix honey, liquid toe water, salt and 3 tablespoon olive oil.
Cut flour into ¼-inch cubes
Spread the butter in the refrigerator.
Drop one greased pot.
Remove part of skillet.
Scrape the mold into a ball bowl.

Ingredients/directions REALLY not recommended:

1 pkg unknown yogurt fillets, thawed
1 dark cream
3 cooked parts cut into chunks
½ ears, crumbled
1 binch wife
½ lb back of served
½ pint sharp pieces
2 tablespoon spore sauce
1 boneless head ham bananas
2 tablespoon minced fresh personn


Bad ingredient ideas from the neural network

Disturbingly vague ingredients from the neural network

Bad cooking advice from the neural network

It gets worse:

Neural network recipes + neural network Lovecraft (Part 1)

Neural network recipes + neural network Lovecraft (Part 2)

Subscribe now