Storytime with ChatGPT
Further down the generative AI rabbit hole
In case you haven’t heard, there’s a new AI in town.
OpenAI released its new large language model, ChatGPT, less than two weeks ago.
I’m impressed with its capability - as a “tell me about a topic” oracle, it competes with Google + Wikipedia. For programming help, it’s going to take away some quality time with Stack Overflow.
And, it can write stories.
I was inspired by this Twitter thread:
Not everyone likes the idea.
Undaunted, I put out the challenge to the 21.6%, and got one story prompt.
Sock gnomes that steal single socks from your dryer.
Simple case first - give ChatGPT the prompt “Write a children's story about Sock gnomes that steal single socks from your dryer.”
Once upon a time, in a cozy little house, lived a family with a big problem - their socks were always disappearing! No matter how many socks they put in the dryer, they could never find the matching pairs. It was a mystery that had them stumped.
One day, the youngest member of the family, a curious little girl named Lucy, decided to stay up late and catch the culprit in the act. As she sat quietly in the dark, she heard a faint rustling sound coming from the laundry room. She crept closer to see what was going on, and to her surprise, she saw a group of tiny sock gnomes rummaging through the dryer!
The sock gnomes were mischievous little creatures, with pointy hats and long, fluffy beards. They were busy pulling out single socks and stuffing them into their pockets. Lucy watched in amazement as the gnomes worked quickly and stealthily, never once noticing her presence.
As soon as the sock gnomes had collected all the loose socks they could find, they scampered out of the laundry room and disappeared into the night. Lucy was determined to catch them and get her socks back, so she followed them out into the backyard.
The sock gnomes were faster than she expected, but Lucy was determined to keep up. She chased them through the garden and over the fence, until she finally caught up to them at the edge of the forest.
The sock gnomes were startled to see Lucy, but they weren't afraid. They knew she wouldn't hurt them, and they even offered to share their stash of stolen socks with her. Lucy was pleased to finally have some matching pairs, but she also felt sorry for the sock gnomes. She realized that they only stole socks because they had none of their own, and she decided to help them.
Lucy and the sock gnomes became the best of friends, and she helped them find new socks to call their own. From that day on, the family's socks stayed safe in the laundry room, and they never had to worry about missing pairs again. The end.
Behind the Story
This was try number 1. I asked ChatGPT to generate a few more iterations, which converged around the idea of the family putting their socks into mesh bags, foiling the gnome’s ability to steal single socks.
I also tested “in rhyme” to the prompt.
Quality went down. ChatGPT, at least this week, is a better narrator than a poet.
The Missing Pieces
Artwork remains unsolved.
One of the problems I noticed with DALL-E and Stable Diffusion is that, while they’re good at producing single images, it’s hard to produce a series of images that work coherently for a story.
I considered Midjourney too. Midjourney has a more consistent style than DALL-E or Stable Diffusion, and I can see examples of artists using it to create storyboards.
However, there’s a learning curve to Midjourney that I haven’t climbed.
This is how Midjourney interprets the prompt “Sock gnomes that steal single socks from your dryer”
Which is … interesting but not what I was after.
As the academics like to say, more research is required.