The Psychology of Prompt Engineering
Mindsets for prompting success.
Wanna become a kickass prompt engineer and work ChatGPT like an “AI whisperer?”
You’ll need to develop a couple of mindsets for the job.
The good thing is that prompt engineering is no rocket science. Instead, mastering ChatGPT is a lot like learning how to talk to a 5-year old…a 5-year-old genius who can sometimes be moody, stubborn, and arrogant, but can also be the sweetest soul on the planet.
You need to coax it, cajole it, and give it what it needs…so that, in turn, it can blow your mind.
We’re learning more and more about AI every day. And so far, here are some handy mindsets to help you thrive as a prompt engineer:
#1 Prompt engineering is not coding.
This is a nod to the coding community: to all the computer engineers, programmers, mathematicians, data analysts, researchers, AI and machine learning scientists, and hardware and software maestros who brought innovation after innovation — to a point where models are beginning to understand natural human language.
It’s because coding has become so successful that non-coders, non-technical people are now able to participate in the revolution.
Lay people with zero programming experience can meaningfully participate.
That’s because prompt engineering is not coding.
Writing code is a very precise, highly exact form of writing. You observe every comma, period, and semi-colon, otherwise, your program will die on the side of the road.
Early on, users treated prompting like coding and were very careful in wording and phrasing the instructions, keenly observing correct grammar, structure, and syntax — afraid that any little misstep can be misconstrued by the language model.
But seriously, this thing can speak over 90 languages and have 1 trillion parameters. Do you think you can trip it up just because you missed a comma or spelled “table” wrong?
You don’t need to lose sleep being hyper-specific or think that only this specifically worded prompt can get such results. This thing breathes synonyms for breakfast. (I mean, can you imagine the corpus of data this thing was trained on?!)
And it’s not that differently-worded prompts will fetch you different results. The response is not different just because you used “big” instead of “large”.
You are bound to get different results even with the same, exact prompt! Try clicking the “Regenerate response” button on ChatGPT and you’ll get a different answer. Try clicking it again and you’ll get yet another answer for the exact, same prompt. (One will be better than the others.)
Don’t treat it like coding.
Treat it like a conversation.
# 2 Prompt Engineering is iterative.
Very rarely do we get exactly what we need with a single go.
Prompt engineering is an iterative process — a trial-and-error, incantation dance in front of a black box. There are no guarantees.
A good prompt engineer will not lose his marbles when he doesn’t get the perfect response after a single prompt. (I know lots of folks who make it a point of pride to get the perfect result with just one prompt, so they load everything into their first prompt.)
But you don’t need to get it perfectly the first time. You won’t! (But hey, if you’re satisfied with generic responses then maybe you can.)
Like a moody child, you nudge it a little here and a little there. You get a feel for what it’s doing and make adjustments as you go. You go through repetitions, corrections, and, most importantly, realizations.
The process is like that of carving a beautiful figure from a large piece of wood. An artist knows that his ballerina is inside that piece of wood. But he doesn’t power through it with a single chop. He patiently carves it little by little, until what he was looking for appears.
Prompt engineering is a conversation.
Don’t worry if you don’t get what you want the first time…or the fourteenth time.
#3 The biggest mental hurdle for the prompt engineer is not believing that it’s possible.
A good prompt engineer will familiarize himself with the full landscape of what ChatGPT and other models can do.
ChatGPT has obscene features, capabilities, and functions. But all that power will lay dormant and kept hidden until unearthed by a set of instructions from a creative prompt engineer.
As such, ChatGPT, powerful as it is, will be limited to the creativity and mindset of the person driving it.
And the type of prompts he can craft, the type of asks, and the instructions he can give, will be limited by his own belief and knowledge of what it can do. He will not ask for things outside the orbit of his belief.
That’s why a good prompt engineer will go on an online expedition and find out what other people have been able to do with it. He will integrate them into his mental model and enrich his prompting powers.
Still, the best prompt engineers are pushing boundaries. Testing limits. Believing that there are yet undiscovered capabilities that ChatGPT is keeping to herself — awaiting a specific set of instructions for it to come to life.
#4 AI is not like you and I.
This one is the antithesis of the previous.
Grotesque and vomit-inducing the powers of ChatGPT may be, it still cannot do everything.
It can fail to solve some Math problems, for example.
AI can only work with inputted data. It’s not all-knowing. And if you let it, it can make for a very frustrating day…or make you foam in the mouth.
Although we are taking ever-longer strides with AI, and almost every day a miracle breakthrough is achieved in the space (there’s new research out that AI can turn thoughts into videos), there are still a lot of things AI can’t do…yet.
This doesn’t deter any prompt engineer from enjoying the thousands of tasks it can do — like creating an original piano piece in the style of Mozart.
The prompt engineer has his work cut out for him. And he has some very exciting rides ahead.
There are others, but remember these 4 mindsets, and you’d do just fine.
Prompt engineering, the role, is really in its infancy, and there’s still so much to discover, and so much to create.
So, onward prompt engineer!
Keep those prompts coming…
