What is Prompt Engineering?
Prompt engineering is the process of designing prompts in order to generate the desired results. A prompt, or input, is the natural language text provided to a large language model that it uses to predict the desired result, or output. Prompts typically take the form of questions or commands. Effective prompt engineering ensures that the output of generative AI models matches the user's intentions and fulfills their needs.
Prompt Engineering for Users
By using prompt engineering to write your prompts carefully and deliberately, you can improve the quality and relevance of your results.
Here are a few tips for fine-tuning your prompts:
Use Precise Language
Generative AI tools rely heavily on context clues to maximize the relevance of their output. Consider what terminology and concepts might be most important for your purposes and include them in your prompt. If you need your results to focus on a particular field of study (such as anthropology, chemistry, or literature), identify it in your prompt. Generally speaking, the more specific the input, the more relevant the output.
Rephrase Your Prompt
Oftentimes, making small changes to the grammar or word choice of your prompt results in slightly different output. It's often a good idea to try different variations of the same prompt to ensure that you receive the most helpful output. You can use a thesaurus to find synonyms of important words or ask the AI to provide you with a list.
Example:
- "Explain the symbolism in the Grimm story 'The Juniper Tree.' / "What kinds of symbolism are present in the fairy tale 'The Juniper Tree' by the Brothers Grimm?" / "Summarize the metaphors in the Grimms' 'The Juniper Tree.'
Break Down Tasks
Generative AI tools sometimes struggle with complex, multilayered tasks, and may provide incorrect information. However, they are often still capable of completing such tasks when they are broken down into simpler steps. You can use language such as "first" and "second" or "step 1" and "step 2" to guide the AI.
Example:
- "First, identify the original language of this text. Second, translate it into German. Third, answer the question in the text in German."
Use Multiple Tools
While many AI tools share similar functions and capabilities, no two are exactly alike. Different AI tools will respond differently to the same prompt, and may provide unique information. Try using the same prompt on different tools and compare each tool's output to identify which is the most helpful.
- Check out the AI Tools for Research page on this guide for a list of possible tools.
Other Resources
Prompt engineering - OpenAI API: A guide for more effective prompt engineering by OpenAI, creators of ChatGPT.
Brex's Prompt Engineering Guide: An overview of prompt engineering by Brex, a financial technology company.