Generative AI tools like ChatGPT have taken the world by storm, challenging our preconceived notions of what AI is capable of and introducing a bevy of new options for businesses to save time, save money, and become more innovative. A few simple experiments can help you see just how much potential there is for these types of tools – but you’ll also be able to see where these tools fall flat.
Thankfully, you can mitigate the worst issues with generative AI and get more value out of these tools by practicing more effective prompt engineering. How do you do it?
The Problems with Generative AI
Casual users may not notice any immediate problems with generative AI, but there are some major issues working underneath the surface. Generative AI tools are extremely predictable and formulaic, relying on repetitive sentence structures and lacking any discernible personality. They can’t give original opinions or offer original ideas, and they’ve been known to give factually incorrect information. On top of that, they don’t have real-time access to the web, so they’re practically useless for generating content on recent developments.
Most of these issues can be addressed through better prompt engineering.
What Is Prompt Engineering?
Prompt engineering with generative AI is the practice of creating better prompts for generative AI tools to work with. Better positioning, better phrasing, and more thoughtful follow-ups can all lead to more accurate and useful generated content.
How to Practice Better Prompt Engineering With AI
How do you do this?
Increase specificity.
One of the easiest steps you can take is increasing specificity. The clearer and more specific your commands are, the better your generated content will conform to your needs. For example, instead of asking for a “tweet announcing a new product,” ask for a “50-character tweet meant to generate excitement for a new type of wall-mounted bookshelf that’s easy to install.”
Provide context.
Context is key. It’s not enough to write a prompt that explains what you want to create; you should also explain why you want to create that content. For example, are you motivating someone to make a purchase? Are you trying to persuade them? Do you want to make people feel a certain emotion when reading this? What is your goal?
Use strategic phrasing.
Your choice of phrasing and your specific words can make a big impact on the content that’s eventually generated. Be mindful of the language you’re using, and be willing to rephrase your prompt to elicit a different response.
Give examples.
Instead of expecting the generative AI tool to understand exactly what you want, feed it examples of what you’d like to see. The more examples you give, the more likely it is to hit the mark.
Use constraints to force ideal generation.
Sometimes, you can milk better content out of a generative AI tool by giving it some functional constraints. For example, you can force it to create content with a specific word count or limited vocabulary.
Obtain alternatives.
Ask to see two or three different variations for your initial prompt. ChatGPT and similar tools can provide you with a stunning variety of alternatives, even without changing or adding to your initial prompt.
Jailbreak when necessary.
There are a variety of strategies for “jailbreaking” ChatGPT and other generative AI tools – in other words, forcing the AI tool to operate outside of its normal constraints. Sometimes, this can help you prod the tool to provide a perspective or phrasing it wouldn’t ordinarily consider.
Give feedback.
After your first prompt, your AI tool of choice will generate some content for you to review. If you’re not happy with that content, or if you think it can be better, provide feedback. Telling the tool you want something longer, shorter, more exciting, less condescending, or anything else will prompt the tool to make another attempt. That’s not always a guarantee of success, but it’s worth trying.
Experiment.
Be willing to experiment. Like with any skill, you’ll become a better prompt engineer with practice and real-time feedback. When you get good or bad responses, ask yourself why and apply those lessons to your future prompts.
Keep learning.
Commit to ongoing learning, as generative AI tools are constantly evolving. Even if you master the art of prompt engineering today, your skills will quickly become obsolete if you don’t evolve with them.
We are all relatively new to generative AI tools, and prompt engineering is a fundamentally new concept. These fundamentals are just the beginning of what will probably shape up to be an entire career field. The sooner you gain mastery over the basics of prompt engineering, the better you can harness these tools to your advantage, in whatever context you deem fit.