By: K Allado-McDowell
Description of the work:
I wrote three books with OpenAI’s Playground, an interface to GPT-3 that made use of the completion API endpoint. In simple terms, that meant I was able to load prompts (such as incomplete sentences or paragraphs) into a textfield and have the model complete them. In contrast to the chat-based LLM interfaces that are common today, this kind of interaction made co-thinking seamless. Human and machine voices merged into the linear flow of prose. Compare this with current conversational interfaces, which always set a two-character scene with distinct, even antagonistic actors. This mode has its own unique dynamics, but I often find myself missing the completion model of human-AI interaction. In fact, to get around this, I vibe-coded a replica of the OpenAI playground using DeepSeek, and had it running in a browser in a couple of hours. It was refreshing to rediscover an exploratory collaborative voice. It got me thinking about the ways that language is reshaped by conversational interfaces. I regularly receive email from people experimenting with AI co-writing, and I love it, but it is often riddled with tropes. You might recognize them: overuse of em-dashes, an always slightly astonished tone, not so much x as y. In my opinion, this abundance of tropes puts a hard limit on AI’s literary usefulness. If we want to get anything good out of AI writing tools we either need to roll our own or learn how to reshape their language as they reshape ours.
Exercise:
Here is an exercise to help you find and calibrate your own voice with a language model.
Exercise:
- Outline an idea for an essay. Describe the general flow of the argument with bullet points. Alternatively, write out a paragraph describing your idea. As you do this, be intentional with your word choice and phrasing. You may need to move quickly and fluidly to get your ideas down, then take an additional editorial pass to structure and shape the language.
- Feed this into your chosen LLM tool. Tell it to generate an essay based on the outline or paragraph.
- Do a close read of the output. Look for metaphors that fall apart, or sentence constructions that go down easy but lack clarity or depth. Be critical of word choices. Take sentences and paragraphs apart to see what’s essential and what can be cut. You may be surprised at what you find.
- Rewrite every word or phrase that doesn’t feel like it comes from you. Let the meaning decompose and re-form. If you have to throw the whole thing out, go ahead, it’s just an experiment.
- Spend some time reflecting on your original idea and what it’s become. Did the LLM make your thoughts clearer? Cleaner? More confused? Has your definition of good writing changed?
- Rewrite the essay without AI. Which version do you prefer?