3.2. How do you formulate prompts to improve the results obtained with GAI?
Specifying the context:
- Provide enough context for the GAI model to understand the task.
- Include relevant details that help guide the response.
Clear instructions:
- Provide clear and direct instructions.
- Avoid ambiguity in the prompt.
Search target:
- Clearly identify the target of the search: what information are you looking for?
- Identify what type of response you expect to receive (descriptive information, instructional information, creative information, etc.).
Keywords:
- Include keywords that are fundamental to the task. These help guide the result.
- Use specific terms that you want the answer to contain.
Constraints and conditions:
- Set clear constraints (if necessary).
- Include specific conditions that the GAI must meet.
Prompt length:
- Check that the length of the prompt is appropriate for the complexity of the task.
- Avoid prompts that are too long or too short.
Specific examples:
- Provide specific examples to illustrate what you are looking for.
- Keep in mind that the examples will help the GAI model to understand the task better.
Direct questions:
- Ask direct questions if you are looking for specific answers.
- Formulate questions clearly.
Iterative feedback:
- Decide whether you need to include instructions on how to adjust the response if it doesn’t meet your expectations.
- Provide examples of how to improve or change the output provided by the GAI model.
Vocabulary and style:
- Indicate the type of vocabulary you want to see in the response or the desired style.
- Specify the particular tone that the tool should adopt.
- Indicate if you want a formal, informal, creative or other response.
Review and adjustment:
- Check the prompt before sending it to ensure there are no errors or misunderstandings.
- Make adjustments and experiment with different reformulations of the prompt.
Jordi Linares, in his article The Art of prompting: Empowering students to harness the power of LLMs (ChatGPT and others) in education, explains another way to generate good results by using more than one prompt. This involves taking a top-down approach: starting by asking for a global view of the topic then progressively diving deeper to get more detail.
Some image-generating GAI models require prompts to use specific syntax. For example, Image Creator (https://www.bing.com/create) requires the following pattern:
adjective + noun + verb + style
For example: futuristic landscape, urban, which captures the sunset, conceptual art.