I will never regard “prompt engineering” as any kind of real engineering. Engineering requires design, it requires an understanding of science so that you can use it invent new technology. It requires accountability for those designs: justifying your designs with regard to cost and function using mathematics, and your math needs to be checked by other engineers to make sure you aren’t making mistakes. Prompt “engineering” requires none of this.

The very term “prompt engineer” is just a marketing term that AI companies are spreading around in their propaganda to try to make their business seem more legit and less like a scam.

With large language models (LLMs), you are basically compressing huge quantities of data into a large, black-box statistical model. You cannot describe the nature of this black box scientifically, its nature changes completely every time more training data is immersed into it. There is no expertise to be gained, no way to say “using these prompt terms the output improves by XX% for the reason that…”

• you cannot gradually learn its nature through experimentation
• the prompts can only be reused for a brief time until the LLM changes
• there is no metric by which anyone could hold you accountable for your prompts once the LLM changes

A prompt “engineer” is just an endless game of guessing and checking, and the results selected will be subject to all the biases of the person running the prompts. It is labor intensive, to be sure, but not in any way that requires design with regard to cost and function. It is NOT engineering.