There’s been a lot of hype (and some opportunistic grifting) around the notion of Prompt Engineering:
“Prompt engineering is a skill of the future! Learn it today, or regret tomorrow. Buy my course to hear 10 multi-use case prompts you can use to get manual work done in minutes!”
But I’ve always been skeptical, and we’re now starting to see that becoming the industry-consensus view. As Ethan Mollick puts it:
“The updated model was delivered in less than two months, and its outputs are indistinguishable from reality. Some notable differences include enhanced coherence, sharpness, and beauty, as well as increased accuracy in response to text prompts.
Most importantly, the model is much better at handling shorter prompts. This is huge news because you no longer need to craft artistically intricate prompts. Just write, “Orange cat reading world’s best AI newsletter,” and voila!
This update also challenges the concept of prompt engineers, individuals who know how to communicate with AI models to receive certain outputs. Engineers with great pick-up lines, if you will. As AI systems continue to learn from feedback and become more familiar with us, they are essentially performing prompt engineering on our behalf.
Why it matters: Knowledge of how to use AI is crucial, but systems are becoming so advanced that the need for specialized expertise might be diminishing.” [my emphasis]
Yes, clarity is hard for most of us, and that’s going to apply to talking to AI algorithms just as it does to other humans. But why did anybody think that AI algorithms that have surprisingly sophisticated and nuanced understanding of language structure would need us to learn a special language just to be able to get across what we want?
[bonus: the stopwatch above was generated with Midjourney. The latest version gets normal clocks right, but the numbers around the edge here are hilariously wrong… ]
I work with leading-edge SAP customers to discuss best practices. Follow me on LinkedIn for more regular updates!