What Is Prompting, Really?
Table of Contents
Generative AI models are improving quickly. We can now generate text, images, video, and audio. But, these models don’t come up with their own ideas. They all require an input that kicks things off, kind of like the suggestion at the beginning of an improv scene.
They need a prompt—that little bit of text that you type into the model—that describes the image you’re trying to create: “draw me a picture of a canoe on a river.”
Or that kicks off the text you want to write: “write me a children’s story about a wizard with French fries for fingers.”
The prompt that you put into the model changes what you get out of it. There are starting to be people out there who like to call themselves Prompt Experts.
Becoming an expert at prompting means that you have learned to harness the power of a generative model, or to uplift the model’s capability. It means that you’ve learned to push the model into interesting places.
Some people spend a lot of time doing this. Many live on Twitter or in Discord channels. They watch how people write prompts, observe the results and then add, remix, experiment, and learn. They share tips, tricks and complex cheat sheets (e.g. The Midjourney Cheat Sheet (V5)) for writing better prompts.
There’s almost an element of magic to this whole idea of prompting, as if the prompts are spells lifted from a dusty spellbook. You cast your magic prompt into the generative AI models, and out of the spirit realm emerges mysterious texts and images.
But prompting isn’t some arcane knowledge to unlock. A prompt is a computer program. Specifically, a prompt is a program that executes within a generative AI model.
When you write a prompt, you’re not doing magic, you’re doing software engineering. And you can make decisions when you’re writing prompts in the same way that you’d make decisions when writing software.
As with any programming language—like Python or Javascript—there’s a language for writing prompts. And gaining experience in that language can help you to get better results from the model.
Prompt language is still a new language. To me, that means the best way to learn is to actually interact with these models and with others that speak the language. Sure, you can sit and read people’s prompts forever. But, actually practising is the only way that you improve. That’s what prompt experts are doing—experimenting, sharing what they discover, then jumping back in to experiment again.
Prompt Prompting #
Prompts can be straightforward (like the wizard story above) or more complex, many people are now experimenting with chains of prompts. Or, prompts that use the outputs generated from previous prompts. You can even use generative AI to create prompts for itself. It’s called Prompt Prompting. This can allow you to prompt a generative text model to write your image prompts for another model. Let’s say you were creating a slide deck about the concept of “The importance of failure”. Here’s how I might approach it with Prompt Prompting:
- Prompt a text model with: “Think of 10 creative ways to talk about failure.”
- Prompt that same text model with: “Create image prompts for each item in the list.”
- Prompt an image model with the 10 prompts I’ve created.
- Prompt the text model with: “Create creative subheadlines that would accompany each image prompt above in a presentation.”
- Build a slide deck using the images and text generated by my prompts.
There’s value in learning how to write great prompts. A generative AI model needs a lot of guidance. When you can learn to give it clear direction, it can create amazing things for you.
So, is prompting the creative skillset of the future? I don’t think so.
Over time, the concept of using complex prompts will likely be replaced by more dialogue-based interactions. Less of a spellbook. More of a conversation.
Because if generative AI is going to be useful to billions of people, it has to be built in a way that billions of people can use it. And, I think it is unlikely that billions of people are going to learn to engineer perfect prompts. Instead, the models themselves will become more responsive to the ways that people interact with them.
Who knows? The future of prompting may not even be writing. What if you could generate a song by uploading a photo? Or a poem by playing an instrument? There’s already progress in this direction–the creative possibilities are endless.
Prompting is a valuable skill. But it’s not necessarily the defining skill of generative models. Because these models are only limited by your creativity, in the same way that a program that runs on a computer is only limited by your creativity.
It’s up to you to be creative.