With AI, it's easy to get lost in the jargon—especially when they become buzzwords that people throw around arbitrarily. One term you may have seen a lot lately is "prompt engineering." Has a prompt engineer popped up in your social media feed yet, telling you how 95% of people are using ChatGPT wrong?
I often see AI influencers and self-proclaimed AI experts calling themselves "prompt engineers" or claiming to practice "prompt engineering." But, is it? What does the term even mean? In my experience, when most people say prompt engineering, they are actually talking about prompt design.
Of course, we are at a very nascent stage in this new AI era, and it's exciting to see new jobs and emerging roles. However, it's crucial to have clear definitions and a shared understanding of these terms. Accurately distinguishing between prompt design and prompt engineering not only improves communication but also helps in hiring the right talent and ensuring everyone in the workplace is on the same page.
In this first installment of a two-part series, we'll clarify the terms 'prompt design' and 'prompt engineering.' Stay tuned for our next article, where we'll dive deeper into the job descriptions and responsibilities of these exciting new AI roles.
Prompt Design
Simply put, prompt design is writing an effective prompt. It's about creating clear and structured instructions, often including specific words, context, input data, and examples, to guide language models (like ChatGPT, Gemini, Claude, et al.) towards the desired output.
Prompt design requires a creative and intuitive understanding of language, psychology, and communication. There are many techniques and styles that a promot designer can use, and every model is unique. For example, people have been able to get better responses by telling a model it will be rewarded, threatening it, or asking it to take a deep breath. My favorite—if you ask a model to respond as if they are a Star Trek character, apparently they are more accurate at math.
Importantly, I'll stress that it's called prompt design because writing an effective prompt is a deliberate process. But how do you know how effective a prompt is? Which technique should you use? What changes will make your output better? This is where prompt engineering comes it.
Prompt Engineering
Prompt engineering is the science of optimizing prompts through rigorous testing and iteration. It's an empirical process that involves developing evaluations, testing prompts against those evaluations, analyzing results, and refining the prompts accordingly. Here’s how it works:
- Evaluation Development: Before writing prompts, prompt engineers create a strong set of evaluation criteria. These benchmarks will establish what a good response looks like.
- Testing: Prompt engineers then test prompts against these criteria, seeing how well the model performs.
- Iteration: Based on the test results, prompts are tweaked and tested again. This cycle repeats until the desired performance is achieved.
The majority of time in prompt engineering is spent on creating robust evaluations and iterating based on the findings, not on writing the prompts themselves.
Summary
So, while prompt design and prompt engineering are closely related, they are not the same. In summary:
- Prompt design focuses on creating detailed and specific instructions to elicit desired responses. It's a blend of creativity and technical know-how.
- Prompt engineering involves the empirical testing and iteration of prompts to optimize their performance. It's more about the process and methodology than just writing prompts.
Think of it this way: if prompt design is about writing the perfect recipe for cooking Jamaican jerk chicken, prompt engineering is about testing that recipe, tasting the chicken, and adjusting the ingredients until you achieve the ideal flavor profile.
So, the next time you hear someone mention "prompt engineering," take a moment to consider whether they're actually referring to the design process or the iterative optimization cycle. By understanding and using these terms correctly, we can have more precise, productive conversations about this exciting frontier of AI development.
In the next article, we'll cover job descriptions and the responsibilities associated with both roles.