Janelle Shane Profile picture
I write https://t.co/WNB8L0jc17. My book "You Look Like a Thing and I Love You" is out now! Research Scientist in optics. she/her. https://t.co/cC9JEun7YH

Feb 7, 2020, 14 tweets

WHAT HAS OCCURRED CANNOT BE UNDONE

I have trained a neural net on a crowdsourced set of vintage jello-centric recipes

I believe this to possibly be the worst recipe-generating algorithm in existence

The training data contained a lot of things.

It contained eel only once. For some reason the AI has decided to use eel a LOT.

It also invents ingredients

Here is another AI eel recipe.

The title is misleading.

Unfortunately that's not a good thing.

When the AI produces recipes like this I feel a mixture of pride and relief.

Then I realize how very very low my standards have fallen.

At least it remembered to pit the peas.

This is one of the AI's more human-seeming recipes (vintage American cooking was EXPERIMENTAL)

Though I have questions about "PG"

Another case where vagueness in an AI recipe is cause for unease

What's interesting is how variable the neural net recipes are in their bizarreness.

This recipe is from the same model & generation settings that produced "Jellied Vegetable Salad" and "Potty Training for a Bunny"

There were some cocktails in the neural net's training data

NO neural net

that is too much extract

One thing the neural net has learned from humans is that it's good to include a story with your recipe.

It is bad at this.

People ask why the neural net generates stuff that wasn't in the recipe dataset. It's because although I finetuned it on recipes, GPT-2 was originally trained on a broad dataset of internet text.

It still remembers how to make Pokemon, and will do so occasionally, unprompted.

The neural net read a LOT of fanfic on the internet during its initial general training, and still remembers it even after training on the jello-centric data.

Except now all its stories center around food.

Today's AI is much closer in brainpower to an earthworm than to a human. It can pattern-match but doesn't understand what it's doing.

This is its attempt to blend in with human recipes

Commercial AI is not significantly smarter than this recipe AI. Humans have just hopefully done a better job of preventing it from making oblivious mistakes.

More about AI mistakes (and dubious recipes) in my book "You Look Like a Thing and I Love You"
youlooklikeathing.com

Share this Scrolly Tale with your friends.

A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.

Keep scrolling