The Art of the Ask: Crafting Effective AI Prompts

When new technologies or innovations hit the hype cycle, more often than not a new, previously unheard-of job role emerges. In recent years, we’ve had blockchain architects, DevOps gurus, cloud solutions strategists and of course, the sexiest job of the 21st century, the data scientist, to pair with the latest “groundbreaking” technologies. And with the explosion of generative AI into the mainstream, we have another job title to get used to.

Introducing the prompt engineer.

The prompt engineer’s remit is pretty simple - get generative AI applications such as ChatGPT or Pencil to do what you want them to do. You don’t even have to have coding or technical knowledge, just the subject matter expertise to know what good looks like and the ability to get the AI to give it to you.

In theory – and I don’t want to start any fights with the prompt engineering community here – anyone can be a prompt engineer, regardless of what they are experts in. They may be skilled designers, copywriters, creatives, strategists or even coders and project managers.

As generative AI continues to evolve, the complexity and importance of crafting precise prompts have become ever more evident, thrusting into the spotlight the delicate balance between the art of human expertise and the science of machine learning.

Related:Power of AI-Driven User Guidance in Digital Adoption

But what does good prompting look like?

Not All Models are the Same

This is fundamentally the most important aspect of prompting - no two generative AI models are the same. Each has its own nuances. Some are better at certain things than others. That could mean image creation or answering math questions - not every model has been created and tuned to do the same things.

Different models have also had different learning journeys. Machine learning principles mean that whoever is inputting into the models, is who the machine is learning from. Different types of people use different models, meaning that knowledge is going to differ from other models’ knowledge. And that’s before you get embroiled in different models’ learned biases and “dirty” data.

That means that the same prompt can’t be used across models. Therefore, a “good” prompter will have an in-depth understanding of the model they are using, the platform it operates on and what outcome is actually possible on that particular model.

But it goes deeper than that – similar models do not mean that similar prompts will get similar outcomes. Take Midjourney vs. Stable Diffusion – you can’t just put in identical prompts and get the same results. There needs to be a considered approach to crafting these instructions and an understanding of how different models interpret them.

Related:A Beacon of Innovation: What is Retrieval Augmented Generation?

In many ways, it is similar to humans and learning styles. Some of us are visual learners, some are auditory and others are kinesthetic. You can’t expect every model to respond in the same way, no matter how similar they are.

This process isn’t just technical but also highly creative, requiring a blend of analytical thinking and imaginative problem-solving. There must be a balance of both strategy and creativity for prompts to provide desired outcomes.

One of the biggest challenges with prompt engineering is the steep learning curve required to master it. Within an organization, the likes of strategic planners to creative designers have varying levels of proficiency in creating effective prompts. This can lead to the quality of AI-generated outputs varying greatly.

And then you have the fact that prompt engineering isn’t a fixed skill. It evolves alongside advancements in AI technology, requiring continuous learning and adaptation. As generative AI models become more advanced in terms of depth, width and complexity, the demands on prompt engineers increase, meaning they must stay updated with the latest developments within AI and refine their techniques accordingly. Prompt engineering is an ongoing process that is dynamic and ever-changing; it is a role in which creativity and scientific precision must constantly evolve and shift in tandem with one another.

Master Prompters Aren’t Made Overnight

Effective prompt engineering isn’t only about achieving accurate and relevant AI outputs - it also serves as a way to optimize the efficiency of the AI development process. By minimizing the need for multiple iterations and refinements, well-crafted prompts can save time and resources, resulting in boosted overall productivity.

Within a business context, prompts aren’t just technical instructions; they are strategic tools that help drive decision-making and creativity. For example, a prompt designed to summarize a document must align with specific business objectives, whether that be spotlighting key insights for a marketing campaign or breaking down complex information for executives. Crafting these prompts requires a layered understanding of both a business and the AI's capabilities. By incorporating negative prompts in a prompt, you are further refining the ask to also exclude elements not required, resulting in even greater efficiencies.

Human Expertise and Scientific Precision – a Persistent Tension

At the core of good prompt writing lies a fundamental tension - one between human expertise and scientific precision. AI models operate on data-driven algorithms, but the prompts that guide them often stem from human intuition, creativity, aesthetics, insights and lived experiences. This dynamic relationship presents both challenges and opportunities, as prompt engineers strive to maximize the potential of AI whilst adhering to technical constraints.

For instance, incorporating visual elements – specifying camera angles or aesthetic styles  – into prompts can amplify the creative output of AI models, most notably in fields like art and design. However, prompt engineers must understand and adapt to the technical capabilities and limitations of AI models they utilize. This ensures that creative ideas can be translated into achievable outcomes, effectively bridging the gap between imagination and execution in AI-driven processes.

Human creativity and the ability to perceive subtle contexts allow prompters to set a baseline for generative AI prompt creation. With human intuition, AI models can be guided in a direction that aligns with human values, understandings and objectives. This human capability allows for a level of flexibility and adaptability that purely data-driven approaches just can’t achieve on their own.

Ultimately, prompting has become increasingly role agnostic, with exceptional outputs being delivered on projects by strategists, project managers and creatives alike. As generative AI continues to evolve, the role of human intuition in prompt engineering will remain vital in ensuring that AI models can deliver meaningful and relevant outcomes. By embracing the art and science of prompt engineering, we can harness the full potential of the technology, driving innovation and achieving new heights of creativity and business efficiency for years to come.

Have you mastered the art of a good prompt?