2 minute read / Mar 2, 2023 /
Pulp Nonfiction : What I Learned Writing a Blog Post with ChatGPT
Over the past week, I put ChatGPT to the test, enlisting it as copy-editor for a forthcoming blog post.
Here was my process:
- I dictated a blog post outline into a Google document
- I prompted ChatGPT to revise the post for clarity & pasted the text
- I copied ChatGPT’s response, edited it some more & consulted ChatGPT for three rounds just like John McPhee
So, does a robot copy-editor work in practice?
ChatGPT builds sentences word-by-word, by starting with one word, then predicting the next one. The robot has ingested more than 500 GB of sentences & calculated the odds that the word after this one is…and then injects it into the sentence.
The robot repeats this process until it completes the sentence or paragraph or tract. Every so often, the robot introduces some randomness in choosing the next word, to pepper its responses with variety.
Without this butterfly-flapping-its wings chaos factor, the narrative spins & stalls, entwining the reader in tautological, repetitive loops.
Like Grammarly, ChatGPT absolves the writer of grammatical labors. I found no errant commas, misspelt words, or grammatical faux pas in its output.
Better than grammar checking, the robot reorganizes sentences, clips others, & trims the text’s length. Mark Twain would have loved this robot.
What’s missing from the experience is a unique voice & creativity : ChatGPT doesn’t yet use words like loaded pistols. No rhetoric. No similes, alliteration, zeugmas, or pleonasms. No statistics to prove a point or ideas from thinkers on the topic without a prompt.
But, ChatGPT proved a useful editor, structuring text, correcting grammar, & helping to hew a raw idea into a polished, final form.
Without a doubt, generative AI robots will produce reams of pulp nonfiction synthesizing coherent re-combinations of sentences found elsewhere on the internet in a singular voice.
The question for startups evaluating automated content production : whether this is enough to stand out in buyers’ minds. For many use cases, uniqueness won’t matter. Product documentation, evergreen content for SEO, canned responses for email.
At least for now, the writer’s role to kindle a spark in the mind of a reader with rhetoric isn’t challenged because novel narration techniques haven’t yet been automated. Whether current generative techniques can get there remains to be seen.