The Structure of a Prompt
Prompting for large language models like Change Agent is a skill that develops over time. As you use the product, you'll find which inputs deliver better outputs. This document provides an overview of best practices for prompting and the key elements of good prompts.
When crafting a prompt, keep the STAC method in mind:
- S: SPECIFIC
Make your prompts specific to get detailed responses. Ambiguous language will give ambiguous responses. - T: TONE
Set the tone or style you want in the response. ie. Professional/persuasive/witty/militant. - A: ASK
Tell Change Agent to ask clarifying questions if it doesn't fully understand the request. - C: COMPETENCE
Specify the competence the AI should use. ie. "Answer as if you are a physics professor with 20 years of teaching experience." Provide relevant details, examples, or data to include in your output.
In the following steps, we'll dive into filling in these blanks and testing prompts to achieve optimal results.
Step 1: Define Your Objective
Identify the purpose of your inquiry, including specific questions, topics, or areas of interest. This will help guide the tone and direction of the response.
Sample prompt:
You will assist me in [briefly describe the topic or question]. Provide an answer based on [key concepts/talking points] for a [target audience demographics/interests]. Include relevant [data/statistics], existing [context/resources]. Write this in a [adjective 1, adjective 2 i.e. “militant”] tone.
More relevant context usually yields a better result. If you found your result lacked specificity, add more detail. Need help filling in context? Ask Change Agent to ask you guiding questions!
Sample guiding prompt:
"You will assist me in [briefly describe the topic or question]. Ask me [three/four] guiding questions that will get us the best result, including key concepts/talking points, target audience demographics/interests, relevant data/statistics, existing context or resources, and response tone."
Step 2: Provide Context and Information
Once you’ve written the outline of your prompt and filled in information, you may upload relevant documents, links, and examples about your topic or organization or goal. This can include text from websites, articles, or previous relevant content.
To load a webpage, press # and paste your link.
To load a document, press + and upload your documents.
Step 3: Edit and Refine Your Prompt
Now, try editing and refining your prompt. Could you narrow the scope of what you are asking? Do you have a certain number of paragraphs in mind? Supplying this directly in your prompt can get a result closer to what you’re intending.
You can even ask Change Agent to refine your prompt. Here is a sample for having Change Agent help you edit your prompt:
Sample editing prompt:
"Edit this prompt for my AI assistant to help my organization, [name], [do xyz objective]. Here is the prompt: [attach prompt here]"
Step 4: Test and Refine
Now you may test your prompt! Ensure all of the information is filled in in the brackets and there are no missing details. With each result, you can assess ways to improve and try adding specificity to your prompt to improve your output.
Try clicking the circle arrow button to regenerate your answer at the bottom of a Change Agent response. Regenerating your answer to review different responses for the same prompt and selecting your favorite is a best practice.
Prompting Pitfalls
- Not enough context: The more context you provide, the better outputs you’ll receive. If you do not provide enough context or relevant information, Change Agent may generate outputs that don't meet your needs.
- Poorly defined goals: Unclear or vague goals can result in outputs that aren't focused or effective.
- Inconsistent tone: Inconsistent tone can make the output seem disjointed or unprofessional.
- Overly broad prompts: Prompts that are too broad can lead to outputs that lack focus or clarity.
- Lack of specificity: Failing to provide specific details, such as target audience, tone, format, or key messaging, can result in outputs that aren't tailored to your needs.
- Ignoring feedback: Not taking the time to review and refine outputs based on feedback can lead to subpar results.
- Over-reliance on automation: Relying too heavily on Change Agent's automation features without reviewing and refining outputs can result in errors or inaccuracies.
- Overly complex prompts: Using overly complex or convoluted language in your prompts can confuse the model and lead to subpar outputs.
- Unclear pronouns: LLMs struggle to understand who pronouns refer to. It is better to repeat proper names and avoid he/she/it/they.