How do I construct an effective prompt?
A prompt is a user-provided input that instructs the selected model on what to do. Engineering an effective prompt is an art and a science, and can be model-specific. You can ask the model to provide information/facts on a topic; to be creative in generating artwork or stories; to provide instructions for performing a task; to compare two or more different concepts; and/or to express an opinion or a perspective on a topic. The quality, accuracy, and relevance of your answer will be directly related to the clarity of your prompt.
To get the most accurate and relevant answer, be clear, specific, and provide context and/or examples.
Be clear. Use precise words to get the desired result - e.g., use “summarize” rather than “provide an overview”.
Be specific. Give specific instructions - e.g., “Summarize the main points of this article into 3-4 bullet points.”
Provide context or examples. Provide examples of what you’re looking for - e.g., “Here are two example summaries. Construct a summary of the article in that style.”
To further shape the results, you can provide specific output instructions, role-based guidance, and even system-level instructions.
Specific output instructions. Provide clear instructions about what you’re looking for as an output - e.g., “Limit the summary to 250 words.”
Role-based guidance. Ask the model to assume a specific persona - e.g., “Construct your response as a sixth grade teacher preparing a lesson for your class.”
System-level instruction. Be clear about the parameters to be used in formulating the response - e.g., “Only cite real sources, and provide links to the sources utilized in your response.”
Didn’t find what you needed? Please reach out to research.computing@dartmouth.edu.