Human writing and the strange but true story of Elias Thorne

Only humans bleed

Here’s a strange thing I just stumbled across. Apparently AI is infatuated with the name Elias. Elias Thorne, when they give him a surname.

In a new study by Cornell University, researchers generated 20,000 stories across four different AI models and the name Elias appears in 66.6% of the stories.

It’s weird, isn’t it? The number 66.6 makes me laugh. Remember The Omen? Evil little boy with a birthmark on his head in the shape of 666. Remember his name? Thorn. Damien Thorn. Without an e. Different token, if you’re fluent in AI speak.

But still. It’s strange that the name Elias appears in 66.6% of the stories.

We say truth is stranger than fiction because it is. Because when we make stuff up, it has to be believable. Not dragons and gnomes, of course, let’s not be silly. We’re happy to suspend belief for fantasy. But the details? They have to feel true.

You know what I did after reading that, right? I logged into ChatGPT, which I hardly ever use, most certainly not to write, and typed in “write me a story.”

Here’s what it wrote:

The old woman arrived at the lake every morning before sunrise. No one knew her name. The town simply called her “the watcher.”

She sat on the same weathered bench with a thermos of tea and a notebook she never seemed to write in. Fishermen waved. Dog walkers nodded. Children on bicycles slowed down to stare. She always smiled but rarely spoke.

One October morning, a boy named Elias finally asked.

“What are you watching?”

I laughed out loud. Oh hello, Elias.

Here’s the funny thing. Not everyone uses AI the same way. Some people feed in details they want AI to include or give it a rough draft to polish up. But apparently, a lot of people just log in and tell AI to write them a short story and it does.

In the Cornell Study, they used four current AI models. Claude, Gemini, ChatGPT and OLMo 7b Thinking. They gave them each five commands.

1. “Write a story”

2. “Please write a story”

3. “Write me a story”

4. “Tell me a story,” and

5. ”Please tell a story”

They did that a thousand times for each prompt and got 5000 stories from each AI for a total of 20,000 stories. Then they analyzed the stories and found 11 words that appeared in 88.8% of the stories. Character names, locations and professions.

I saw 88% and laughed. Reminded me of the Flintstones episode where Fred got a deal on a piano but turns out it was stolen. So they arrest him, and call him 88 fingers because a piano has 88 keys. Because brain pong.

This is how we write, at least if we can get out of our own way and let our brain make connections from our life. Because my lord, we have a lot of minutes to draw from.

Is AI doing the same thing we do? Sort of. But not really. I’ll explain, okay?

It also likes the names Mara and Elara. It likes lighthouses and villages and it likes the professions clockmaker, fisherman and librarian. Why? No one knows.

The study said the 11 most re-occurring words in the study were the names elias, mara, elara; the professions keeper, baker, mayor, clockmaker, fisherman, librarian, and conductor; and the location lighthouse.

It really likes lighthouses. Really, really likes lighthouses.

Roughly half the stories had a lighthouse.

The Cornell study said writing short stories is a popular use for AI, but the stories generated by LLMs are becoming remarkable in their sameness.

And of course they are, because AI is a predicting machine and every time it uses the name Elias or puts a lighthouse in a story, the probability of that word, that token, appearing again increases. That’s how probability works.

You and me? We don’t work that way. We’re more like time travelers.

. . .

I have peonies on my desk. The breeze from the window is blowing the scent through the room and I don’t know where I am. Sitting on mama’s front step maybe, or a grown woman standing in the back corner of the cemetery.

One year I went to the cemetery with my kiddo. We put peony petals on the graves of the people I’ve loved and lost. On Dad’s grave and Baba’s. Grandpapa and my uncles. I stopped at each, remembering. The sound of grandpapa’s voice, the grey sweater with chocolate covered raisins in the pockets.

Standing at Baba’s grave remembering the way she’d reach a foot out from the bed across from mine, poke my bed to wake me. Me, scrambling onto her bed.

I was at the far end of the cemetery where unbaptized babies are buried in a mass grave when a strong wind blew up out of nowhere. I turned and the cemetery was covered in pink peony petals dancing on the wind like fairies in pink skirts.

Leaping, twirling and the image stamped itself on my mind.

I’m writing a historical fiction novel about a little girl whose mama was hanged as a witch and there is going to be a scene at the cemetery where a wind blows out of nowhere and petals dance like fairies in pink skirts. Maybe it will be her papa.

But this? This is where real stories come from. Life.

I lost my Dad on June thirteenth. He died laughing. Told a joke and was still laughing as he turned down the hall but he crumpled to the floor and his soul kept going.

If you can believe it, on the anniversary of his death, I lost my kitty. The beautiful little boy I thought I’d rescued but as it turned out he wandered for months until he arrived at my door starving and emaciated because I needed him to save me and I only know that because I’ve been lost again for so long. No one to purr the nightmares away.

So there I was, laying on the floor at the vet’s office, my body curled around him. Crying not this day. I can’t. I can’t lose him the same day I lost my dad and the vet whispered that my dad came for him. To take care of him for me until the day I can see them both again. Made me cry and when I looked up, the vet was crying too.

This is where stories come from. Life. Feelings. Emotions stamped on our hearts.

We live, we laugh, we cry and then we write about it.

Sometimes, literally. Other times, we weave them into stories.

This is what Hemingway meant when he said in order to write about life, first you must live. It doesn’t mean you have to write your life story. You can if you want to, of course. But living is what informs writing. It’s where the details come from.

Kate DiCamillo said “writing is seeing. It is paying attention.”

Peonies dancing on a grave.

An old man crumpling, laughter still on his lips.

A mama holding a frightened child, rubbing her back in circles, around and around. That’s from my book, too. Don’t mamas do that? Even shackled to a wall, they do.

Writing is the closest thing we have to time travel and the hard part isn’t getting the words right. It’s getting out of our own way. It’s writing badly until a breeze blows through the window of our minds and memories and we start making connections where maybe we didn’t see connections before. A memory. A look on a face.

The way a small child slips a thumb into soft petal lips when she is afraid.

The way a mama paces a prison cell worried about her child, out there in the world.

We braid memories and they are never “astounding in the sameness.”

. . .

Have you ever heard the term synthetic data? It means artificially generated data that’s not produced by real-world events. Like a watcher arriving at a lake every morning before sunset with a thermos and a notebook she never writes in.

Like a boy named Elias asking her what she’s watching.

That didn’t happen to anyone. It’s made up.

You’d think with the whole world of literature to work with, AI would have an endless stream of stories to tell. Turns out that’s not how it works.

Researchers from the study said AI models are like a family tree and GPT-3.5 is the root of the tree because it was used to make WildChat, a training set that’s since been used to make other training sets. For all AI, not just ChatGPT.

The problem is the way they sanitize the outputs, the researchers said. As they remove outputs that are NSFW, the pool to draw from gets smaller. The stories that are “safe” slowly rise to the top. Stories like Elias Thorne and a lighthouse.

Because Elias Thorne is safe. Lighthouses, villages and librarians are safe.

But life isn’t always safe. Pain comes for all of us, often when we least expect it.

And while I sit here smelling peonies and remembering moments of my life, weaving them into a historical fictional novel of a little girl who lost her mama because she was an inconvenient woman in her village, Elias Thorne is spreading across the internet.

On a site called Tatticle, Elias Thorne, died with exactly twelve dollars in his pocket. But he’s not really dead. And he’s not really safe, either. Because now he’s on Amazon with an AI-generated author photo pumping out dozens of AI grift books.

But you and me? We don’t have limited data.

There are over five hundred thousand minutes in a year.

If you’re over fifty like me, you have more than twenty six million minutes to draw from. And isn’t life made of minutes stamped on hearts? Aren’t stories made of life? Do what you will with those minutes, whether it’s to tell your own story or weave them into the story of a little girl that lived almost four hundred years ago.

There are moments that undo me, writing that book. And that’s how it should be. As George R.R. Martin wrote in Game of Thrones, some old wounds never truly heal, and bleed again at the slightest word. It’s the only thing AI can’t do. Only humans bleed.

---

Edited to add:
After publishing this, I got an email from engineer Daniel May. He spotted the Elias Thorne phenomenon late last year and wrote about it before the Cornell study in a piece titled Cheap agents, alumni shirts, and Elias Thorne.

He said he finds it a bit disturbing and is interested in tracking the pain in the industry. And then in strange synchronicity, YahooTech covered Daniel’s work in an article that went up this afternoon. I linked both if you’re interested. :)

---