The Word Smiths
Elizabeth heard the alarm go off. She groaned as the blinds opened and cool light flooded her bedroom.
Her computer turned on automatically. The bright flickering of the screen drew her out of bed though she was still exhausted. Yet she had a job to do.
She rubbed her eyes and placed her fingers carefully on the keyboard. Then she began to type.
Her mind, still waking up, had no idea what the flow of letters were supposed to be from one to the next. Yet still words appeared in front of her. Perhaps her subconscious was trying to communicate with her.
For a brief moment her eyes glanced to the toolbar at the bottom of the page. "900 words remaining" it stated. Then she could get her first break.
She mumbled to herself as her fingers continued to tap on their own. She had to buy a blindfold one of these days, she reminded herself. Typos didn't matter, as long as there was quantity.
A calendar notification appeared. Coffee and lunch? Ah, it was the first Tuesday. That was when she met with the other workers in-person. At two years, she was already the senior-most wordsmith at the company.
She yawned. Now only 500 words were left. Her talent at speed typing had gotten her this job. Each day was the same, but at least it was stable work. The world would always have a need for word smithing.
Words continued to flow from her fingers, a mix of poetry and banality. She tried her best to put flourishes where possible, her supervisor liked that, especially since it never took more time.
Two hundred. Then one hundred. Then done. A final period at the end of the word gave her the first completed assignment.
Her stomach growled, as she hadn't eaten breakfast yet.
As she looked up again, she saw her words had already disappeared. They were already being used to train the AI.
I was inspired by this tweet, which I had saved a while back aiming to come back to this idea. Large language models are impressive but not ready yet. They need much more training data. Will they ever have enough? Are there even enough words in the world yet? Maybe we’ll need people to create more training data.
This is a big problem in English, but I can imagine that smaller, less spoken languages may never work perfectly.