How Not to Write: Love, ChatGPT
My husband’s a great man — smart, funny and a terrific father — but a romantic poet he’s not. So when he sent me an email with the subject line “Love Letter,” I opened it with some skepticism.
My Dearest Jamie,
You are my best friend, my soulmate, my everything. I am so grateful for your unwavering love and support. Your smile brightens my day, your laughter fills me with joy, and your touch is like a warm embrace that I never want to let go …
I winced. The writing was trite, cliched, devoid of specifics and written in a voice that was oddly formal yet overly saccharine. Basically the complete opposite of my husband.
You see where I’m going with this. The email closed, “Yours Always, ChatGPT.”
I laughed with relief. Partly because I wasn’t going to have to pretend to like this painful Ode to Moi. But mostly because I realized AI wouldn’t be coming for my job as a fiction writer. (At least, not yet.)
That said, in the following weeks an uncomfortable truth descended: The realization ChatGPT’s bad writing is simply an extreme version of my worst — from those early copycat novels I never finished in high school to those sleep-slurred ideas I email myself when I suddenly wake up in the middle of the night, to those lazy “xxxxdescription goes herexxxxxx” tags I employ when I’m drawing a blank.
And if you’ve played around with any of the chat or art generators, you’ve also learned you can significantly improve what they produce by giving them more direction – sensory details, unique action, all done in the style of … whomever (all of whom, by the way, deserve money for the use of their art).
So it goes with our own work. Which brings me to my point. Rather than fretting about the inevitable rise of our robot overlords, or even laughing about how awful AI fiction currently is, we should learn from it.
Next time you’re writing and frustrated, be gentle with yourself. Imagine you’re a baby AI. It’s not that you suck, you just need to cycle through a few more iterations. Cut those cliches. Choose more specific verbs and anecdotes. Experiment with voice. Take risks. Avoid formulas. Get personal. Get weird.
AI will only improve, but at their core, these language generators are of a predictive nature. They spit back what’s statistically most probable.
So do the opposite. Surprise the reader. Better yet, surprise yourself.