What I learnt about designing effective prompts

This week, I was busy studying prompt engineering. The task also coincides with something I am doing on the personal front… more later. The task lent me a window to pare into the world where AI and writing coincide. And I couldn’t help putting together a handy list for reference.

This post is about what I found about the question, “How does one create an effective prompt for AI?”

A cue for priming the AI: Define the task or your problem at hand. For example, “Looking for shoes.”A word of appreciation: Instruct the AI about the appropriate style and tone. Your prompt should also include instructions on how the resultant content can be made easy to read, comprehend, and apply. For example, “Preferably for non-formal, casual wear.”A word of caution: Share information on handling error cases. You provided your expectations. Now, supply what the AI assistant must do if it can’t produce the resulting information. In this case, such a resultant scenario is unlikely. So, let us rephrase the exception. For example, “You can exclude black because I dislike the color.”A reminder about relevance: Contextual relevance is of paramount importance. Set the context for your your AI assistant to consider your point of view while producing results and responses. For example, “Something that I can use while traveling or commuting through the city.”A hint about desires: Instruct your AI on the format for the desired output. Supply instructions that help your AI assistant to produce actionable information. For example, “I’d prefer sneakers over Oxford-style shoes.”

Your questions are determined based on your need. And, your need is based on a choice or decision you wish to make. So a good prompt should produce results and responses that help you decide more effectively and efficiently. For easier reference, concentrate your prompt on three points: why, who, and how.

Does it mean you load the prompt with everything that you can define? No, the more elaborate prompt confuses your AI assistant. Just like our users, even AI assistants rely on just the right context for the right responses. I’ve found that adding up to three contextual references helps in most cases. Your prompt can look like any one of the following options:

“I am looking for shoes for myself. Preferably for casual wear. Something that I can wear while commuting to my workplace or coffee shop.”“I am looking for shoes for personal use. I prefer sneakers over Oxford-style shoes. So, exclude black.”“Looking for shoes for gifting. Preferably sneakers. Or anything casual.”

I’ve come to understand that training AI assistants for the right responses is like training your kids. You’ve got to get the right mix of input to get any desired output. And, much like parenting, prompting isn’t easy.