Without a doubt, AI and ChatGPT have forever transformed the way artists and writers create. At this very moment, screenwriters are striking over the use of AI in film and television. And to be sure, AI is a powerful tool, able to churn out truckloads of workable stories and ideas. Yet just as many times, AI can generate some pretty insane snippets of text. Called “hallucinations” these hint at the way AI thinks and interacts with the world. As writers, we can exploit AI mistakes to jumpstart our own creative projects.
Thinking in Tokens
AI and ChatGPT do not think or process language the way humans do. In the short run, this makes them able to crunch data in a matter of seconds. These chat bots feed on vast amounts of text and then spurt out a pretty good answer. But in the end, they see the world as bits of text called tokens.
AI does not generate answers letter by letter or even word by word. Instead, they use something called a token, which is a snippet of a word. Mostly because there are too many words to parse but too few letters to really make meaning. Tokens are the middle ground.
Ask it to understand a nonsense word like overpythonize and it breaks it into manageable tokens (over + python + ize).
Blurry Hallucinations
AI is an amazing bullshitter. It will always give you an answer and in a confident way, despite the fact that the answer might make no sense. Ask what Napoleon thought about hamburgers and it spits out that the French leader thought hamburgers were the food of the gods.
Again, this comes from the difference in the way AI processes information. Computer programmers set up chatbots to deal with massive amounts of information in a short time. This means they have to cut some corners in deference to speed. In computer speak this is called lossy compression.
For example, let’s say you feed ChatGPT a prompt about skateboarding: “Write about how to skateboard in the style of a romantic poet.” The AI looks at the two different ideas, skateboarding and romantic poet, and it interpolates the missing information between the two. Here’s the response:
Oh, how my heart doth yearn to glide
Upon a board of four-wheeled pride
To feel the wind, to feel the rush
And hear the wheels’ melodious hush
ChatGPT is amazingly good at this sort of blurring or melding of styles. Us humans enjoy the output to no end.
Yet this surface playfulness hints at a deeper flaw in the way AI looks at the world. Yann LeCun, a French expert in computer learning, says, “Large language models have no idea of the underlying reality that language describes. Those systems generate text that sounds fine, grammatically, semantically, but they don’t really have some sort of objective other than just satisfying statistical consistency with the prompt.”
ChatGPT is the Ultimate Cut Up Tool
In the early twentieth century, surrealist artists developed a technique called cup-up, where bits of writing are cut apart (sometimes into token-like bits) and rearranged to create something new.
One advocate of cut-ups was writer William S. Burroughs. He saw it as a way to break away from traditional narratives and linear thinking. This approach could reveal hidden truths and dip into the subconscious mind of the author.
Sometimes AI chatbots will generate some truly strange word combinations: “the moon was truly mother-of-pearl, the white of the sea, rubbed smooth by the groins of drowned brides.” Only someone like Burroughs would use the sentence as is. Yet the odd word juxtaposition creates a frictive force in our brains.
The responses generated by AI are often filled with cliches (any prompt about fantasy is likely to involve swords). And sometimes it delivers absurd wordplay like “the groins of drowned brides.” Yet the surrealists and beat poets who used cut ups knew that being a creative person, or a writer was not just generating words, but evaluating whether or not, the content has impact.
We can utilize the strange and bizarre mistakes generated by AI to spur our own creativity. The chain of thought initiated by the “going of drowned brides” could lead to some innovative creative writing.
The “hallucinations” of AI can unlock our own unconscious. They serve the initial intent of the surrealist cut ups. Yes, Chatbots can generate massive amounts of text. But it’s up to us, the thinking humans, to evaluate if that text is any good and how to use it.
Tim Kane