How to Engineer AI Prompts in Service of Your Business
In its current state, artificial intelligence takes whatever you tell it very literally. As such, it is very easy to misdirect it into digital rabbit holes… which is the last thing you want, when time is very much money to your business. This is precisely why it is so crucial that we become adept at properly prompting the AI models we use. Too many hallucinations (responses that share inaccurate or unreliable information) simply waste time and money, but the better the prompt, the less prone the AI will be to hallucinate.
Let’s go over some of the best practices to keep in mind as you draft your prompts.
A Better Prompt is Possible Through the RISEN Framework
If there’s anything that AI depends on, it is a clear set of instructions. As such, sticking to a rubric that is designed to provide maximum clarity and specificity is important.
The RISEN framework is a very effective way to do just that. By following it, you can better collaborate with the platform you’re using and coax out better, more contextual responses.
The RISEN Framework, Broken Down
Like many frameworks, RISEN is named for the acronym that outlines its steps:
“R” is for Role – Whose perspective do you want the AI to speak from? After all, the response you’d get from a stand-up comedian will be much different than one you’d receive from a data scientist or a marketing professional. Identifying “who” the AI should emulate is the first step toward getting the response you need.
“I” is for Instructions – What do you want the AI to accomplish as you give it its task? While this is somewhat self-explanatory, we’ll dig into the minutiae in the next steps. In essence, it is important to provide more detailed instructions for the machine to follow, simply so it has enough to reference.
“S” is for Specifics – Keep in mind that you have an idea of what you want the AI to deliver. The AI does not. Therefore, it is important that you include these requirements in your prompts. This is particularly important when industry standards such as HIPAA (the Health Insurance Portability and Accountability Act) and NIST (the National Institute of Standards and Technology) apply. Provide the AI with as much detail as possible so it can incorporate all guidelines and specifics into its processes.
“E” is for Example – Give the AI something to reference as it configures its response. Point it towards something you want it to emulate, like a preexisting report or an example that fits well. If you own content that matches this, pointing it to that is ideal.
“N” is for Negative Constraints – It is just as important that you set guidelines in terms of what you don’t want the AI to do. Make sure it speaks appropriately to the audience the content will be directed towards.
Above All Else, You Need to Provide Context
Again, the AI will only know what you tell it. Identify your audience, explain what you are looking for, give it examples of your voice for it to emulate, and critically, work with it over and over and over again.
That said, you should never, ever, provide public AI tools with sensitive or proprietary data. These platforms are built on information sharing, so anything you provide to them could potentially be used to build responses to someone else.
What If I Want to Adjust the Tone?
Check to see if your AI tool of choice has settings that allow you to change the “temperature,” which is essentially how the AI approaches your prompt. High temperatures lead to more creative responses, while low temperatures lead to more factual responses.
Get More Out of Your Prompts and Give More to Your Business
Milk your efforts just that much more by allowing AI to augment your outcomes. We’re here to help you work with technologies more proactively. Give White Mountain IT Services a call at (603) 889-0800 to learn more.