A year ago, generative AI was a curiosity at best for the general public. Now, anyone who knows what they want and has an internet connection can use services such as ChatGPT, DALL-E and more to make their thoughts a reality. Yet the output of these services is only as good as what’s put in—namely, what you request the service to create for you. Not all inputs are created equal—asking ChatGPT to “Write a blog post on how cryptocurrency works for beginners” versus phrasing the request as follows: “Write a short article on how cryptocurrency works from a technological standpoint. Your audience consists of people who do not understand the underlying technology. Use simple terminology.”
The need for minimizing time spent refining the outputs generative AI produce has given rise to a new skill known as prompt engineering. Prompts are the foundation that generative AI builds its output on, making their comprehension by the service a top priority. For newcomers to prompt engineering, it can be a daunting task to create an effective prompt that will produce what you want the first time. Below are several tips to help you understand how to build better prompts.
Keep It Simple
When it comes to prompt engineering, simplicity is key. While it may be tempting to include complex and technical language, it’s important to keep in mind that the service does not think like a human and have the ability to reason as we do. Opt for simpler language that is easy to understand, and avoid using jargon or technical terms unless they are absolutely necessary. If you require an output with more complex language, specify that in the prompt.
Be Specific
In addition to simplicity, specificity is also important in creating effective prompts. Vague or ambiguous prompts can lead to confusion and ineffective responses. Instead, use clear language and action words relating to the subject at hand. This will help the AI generate more accurate and relevant responses. Give the AI context, such as telling it that it is an accomplished technical editor named Winry, or Alphonse, an expert resume writer.
Leverage Memory
Services like ChatGPT can remember information previously introduced. If you will be generating numerous outputs based on a common subject or topic, it may be prudent for you to feed the service information you want it to utilize or represent in its output. This can be as simple as including several lines of information in your prompt, or submitting prompts before your true outputs that instruct the service to remember what you type in the rest of the prompt.
Test, Test, Test
Finally, test your prompts extensively. Start with a small set of prompts and test them thoroughly to see how they perform. Even though generative AI is not sentient, you can compare the output versus your input and adjust your future prompts with those discrepancies in mind. If the resultant outputs are too wordy, consider adding a request to summarize the response in one paragraph or several sentences. No one creates a flawless prompt the first time, and this is still a new skill. The old adage of ‘practice makes perfect’ has never been more more applicable!
Leave a Reply