AI Whisperer: Job of the Future?
I was thinking about this idea that a high-paying job of the future is going be prompt engineer.
I was thinking about this idea that a high-paying job of the future is going be prompt engineer. An “AI whisperer” if you like. This person is able to talk to the AI in a way that consistently provides the output they are looking for. One place where you can see this in action is in the field of AI generate art. Maybe you can say, “I want a picture of an astronaut riding a horse on the moon” and get something fun and cool.
But what if you wanted an 80-year-old female astronaut in an old NASA suit, her face clearly visible behind the faceplate, riding a purple horse with a kind of futuristic armor that generates oxygen and protects his body (never mind that the astronaut is wearing a suit), with a high-rise Moon base in the distance.

Astronaut riding a purple horse, by Marcel Gagné, created with Stable Diffusion
Okay, it’s close, but not exactly what I was looking for. This, pundits will tell you, is where the AI whisperer (or prompt engineer) comes into play.
Except, I’m not buying it.
Fact is, I can write a pretty great paragraph that describes a scene with (or without) a central character. I can describe the landscape, in some detail, and impress a certain atmosphere in that description. However, turning that paragraph into a prompt that generates what I want, what I have already created, is not much different than a programmer writing a piece of code to add up two numbers.
What has made digital assistants like Google Assistant or Amazon’s Alexa so cool is that I didn’t need to write out something long and esoteric just to have it play some light jazz for me, or to tell me about the weather for tomorrow.
The real promise of AI is not to force a new generation into learning the finer points of crafting prompts, but to understand us and to collaborate with us in a way that is natural. AI that understands what we want is becoming more and more commonplace, with modern AI easily passing complex feats of understanding like the Winograd test. Generate AI systems have been hacked to accept an overarching request, or prompt, and then generate its own prompts, as needed, in order to achieve that goal. Look into