Imagine you are talking to an architect about a new house you want built, and the only thing you tell them is “I want you to build me a house”. The architect won’t have any idea what specifications you want, what your needs are, or what purpose the house will serve, so they build you the most generic house they possibly can.
Now imagine instead of just telling them to build you a house, you tell them you want a 3-bath, 4-bedroom house designed after Frank Lloyd Wright’s Fallingwater house and it will be used for entertaining your entire extended family during the holidays. With this information, the architect will be able to build you a house that is meant for you, not just a generic house that could be for anyone.
How does architecture relate to artificial intelligence?
Using AI models like Chat-GPT and Gemini works in a similar way. If you tell an AI model to write you a recipe for soup, it’s going to give you a random soup recipe that may or may not be helpful for you. However, if you tell it to write you a recipe for dairy-free tomato basil soup using all-organic ingredients, it’s going to give you exactly that.
The process of using specific instructions for AI models to get the results you need is called prompt engineering. While some prompt engineering can be easy to come up with, there are many specifications you can give artificial intelligence to achieve the output you desire.
Example of prompt engineering techniques:
Role: Who do you want the AI to be “playing”?
Depending on the type of results you are looking for, it can be helpful to identify the role you want the AI to play. Examples of roles could be a teacher, a lawyer, a parent, or an alien from a galaxy far, far away.
Tone: Do you want the response to sound a specific way?
Giving a specific example of a tone you want the response to be in can help shape the voice of your results. Tones like formal/informal, comedic, persuasive, or informative are good examples of ways to guide your AI results.
Audience: Who is the response for?
Another way to help get the best results is to specify what the audience for the request is. If you are asking AI to write you an academic journal, you can include that the audience is higher education students and professors. If you want a poem fit for 1st-graders, informing the AI of those details is going to reduce the edits you will need to make in your results.
Specific Request: What exactly are you looking for?
This may seem obvious to some, but it is surprisingly easy to overlook. AI models like Chat-GPT thrive on specific information; if you want a blog post written, specify the word count you are looking for, how many paragraphs you want, what the blog sections should be about, or whatever else is relevant to your desired outcome.
Purpose: What is the goal of the result?
Identifying what you are hoping to achieve from the request is another way to help guide the AI in the right direction. If your goal is to sell something or generate leads for your business, your result will be more persuasive. Similarly, if your goal is to educate, the results will be more informative.
Extras
Other ways to help engineer your AI prompts are providing specific examples of what you are looking for, asking for adjustments when you want something changed, and adding any relevant requirements like “must include the word purple in every sentence.”
Building a Result
You can easily ask Chat-GPT to give you a list of silly pet names and synonyms for cinnamons without having to add any prompt engineering. In fact, most everyday AI requests won’t require many extra details. However, if your request is longer or more complex, it’s good to be mindful of the ways that you can help AI help you. Chat-GPT and similar models aren’t able to customize your results without you telling it what you want to receive. AI is similar to real humans in that way: If you don’t tell them, they aren’t going to know.
Use these prompt engineering techniques next time you ask AI to write you something and see how your different inputs change your results!
