NOTE: The experiment here was to use prompt engineering to write an article about prompt engineering. The LLM I used was Claude 2. I knew which articles I wanted to use, and the main point I wanted to make. The end result was just about perfect. Here it is in its finished form.
Prompt engineering has become a bit of a buzzword in AI circles lately. For those unfamiliar, it refers to the careful construction of prompts fed into large language models like ChatGPT to shape the desired response.
With the right prompts, these AIs can generate remarkably human-like text on virtually any topic.
But as AI researcher Oguz A. Acar noted in a Harvard Business Review article called “AI Prompt Engineering Isn’t the Future, “the prominence of prompt engineering may be fleeting for several reasons.”
He points out that future AI will become more adept at understanding natural language, reducing the need for complex prompts.
New models like GPT-4 already show promise on this front. And prompts are heavily dependent on the specific algorithm, limiting their broad usefulness over time.
Some best practices for effective prompting include being as specific as possible, providing contextual information, and giving examples to the AI model.
As Google’s Cloud blog discussed, details like specifying you need a recipe for 50 people rather than just “blueberry muffins” make prompts clearer. Context about the situation and desired tone also guides the model.
In many ways, though, prompt engineering is essentially what good writers have been doing for ages—using thoughtful language to articulate ideas and elicit responses from readers.
We’ve always crafted prompts, even if we didn’t call them that. A compelling headline prompts further reading. An insightful op-ed prompts discussion. A well-posed question prompts thoughtful replies.
This ability to carefully shape language to drive engagement is an innately human skill. AI may get better at constructing its own prompts through advances like few-shot learning, but it lacks the creativity and empathy we inherently possess.
We understand culture, context, voice— nuances no algorithm can yet grasp.
So while AI prompts may evolve and even outlive their usefulness, human prompting through writing will remain an impactful act. Our words have power. And the prompts we engineer, in all forms, can change minds, inspire action and connect us more deeply.
Rather than fearing prompt engineering’s demise, we should have confidence that thoughtful writing and messaging will continue influencing people for good.
Our prompts’ destiny lies not in algorithms, but within our own ability to craft language that profoundly impacts human hearts and minds.
This is a great end result that conveys my intention. Here’s how I got there.
My first prompt was this:
Claude hallucinated immediately and got my name wrong, so right out of the gate we had to address that issue. I corrected Claude and we moved on.
Once we were back on track, I wanted to add some additional information from Google’s Cloud blog, so I fed Claude the information I wanted it to have.
Claude regenerated a new article based on these instructions. My last step was to add the links.
As you can see, giving Claude the right prompts, and feeding it the information you want it to synthesize can lead to a powerful result.