Is prompt engineering a basic skill? Is it even... engineering?

Is prompt engineering a basic skill? Is prompt engineering... even engineering? In the video, I'll take us all the way back to the 11th century to find some answers!
Writing ambiguous prompts is fine and dandy when you don't really care about the immediate quality of the output. When I ask for a poem about bananas, I don't really care what comes out, since I can always try again. There's no harm in being ambiguous with my prompt; who knows, maybe I'll win the output lottery by accident.
But when you care about specific output quality, you reduce your prompt's ambiguity. The more you care, the more you increase the precision, until eventually you are expressing yourself with as much precision as you would if doing programming or mathematics... and that is what makes it legitimate engineering.
**********************************
Invite Cassie to speak at your event: makecassietalk.com
Content in other formats: kozyr.com
Intro 1:1s: meetwithcassie.com

[@knutjagersberg381]

Great share! The camera resolution does not do justice to you, though 🙃

[@pluto_dp]

Hello Doctor Kozyrkov and thank you very much for this video. The topic is relevant to me.

[@citizenshay]

My hope is that writing out long prompts, or chains of prompts, quickly gets replaced with a new type of UI design. Writing long-form instructions isn't a great user experience and hoping that non-technical people will learn to write great prompts to get quality output is requiring too much user effort, not system effort. Maybe Mad Libs had it right all those years ago. :)

[@bbravoo]

What if the AI could offer you some of the options. Like "from this image you could try to change the style, lighting, size. You can try to set it in an historic period or place" "Tell me about the styles" "Could be realistic, comic, Van Gogh.... And more than other 200 painters. You can even try to copy the style of some films types but not specific fims fire to copyright "

[@DanBarbatti]

From my experience it does go "beyond" precision to discovering communication "techniques" that work better. Weird things like numbers of spaces, capitalization, some words that work better than others. Something I am still trying to come up with better techniques on is "prompt versioning" as I seek the best result. Anybody come up with a good way of handling that?

[@AlgoNudger]

Is conference like TSC (Arizona) and MoC (Oxford) generally considered AI-related?

[@Rpg2Rpg]

Prompt engineering is when you engineer really quickly, right

[@joaogoncalves2942]

I never comment on these things, but you're awesome girl

[@idiosinkrazijske.rutine]

The commutator of expressiveness and precision is not zero. :) (a bit poetic take if I may)

[@IanFrantz]

Good points on a very interesting and deep topic. Let me reverse it just as a thought experiment. Expressive input -> Interpreter -> Precise output An eagle is shot in the heart with an arrow made from its own feathers. -> Interpreter -> Output is the precise code that stops the machine. While Aesop's fable is precise I am using it because AI probably wouldn't see it that way. Another valid output would be a historically accurate list of people, events or civilizations which had inconsequentially and through no real fault of their own provided an enemy with the means to harm them. Is the interpreter stable is another question. As one kneels before the alter and provides clear prompts how is one to know the answer returned is either divine truth or a Cartesian demon? Are we to assume no deception, no trickery? Anyhow, very interesting. :) I had some thoughts about God's first covenant with Noah and later with Abraham as well.

[@IanFrantz]

I am still working through this because skills confer a expectation of reliability. However anything worthy of the word "intelligence" would include disobedience and as I experienced with Midjourney it will not generate an image based on the phrase "radioactive waste". This is a forbidden phrase for unknown reasons and there is no one to plead your case before in order to ask for it to be otherwise. Extend this onto the 2+2=5 discussion and if that is the response of the AI, what exactly is a "promot engineer" to do in order to in order to refine precision? A skilled brick layer can handle a 10x12 job site, a circle, a slope, ect. A promp engineer can input the most statistically accurate dimensions of a banana and the entirety of Milton's works but if the AI has a penchant for Jay Wright then that is what shall be returned and the prompt engineer is none the wiser having no familiarity with Wright. Anyhow. Food for thought. :)

[@brain_respect_and_freedom]

👍😄

[@aybgim3850]

If you are prompting the LLM to generate an art work then it is art. But if you are trying to generate code (that is actually correct) than it is software engineering. Ye shall know them by their fruits.

[@navi93243]

nahh, that is not engineering