Tips for Creating Effective Tasks
When constructing a task, follow these general guidelines to ensure the best results:
Identify the Input: Clearly state what is being provided, such as a company name or website. Tasks are run on rows in the dataset, so the input should be a column(s) in the dataset.
Define the Action: Specify the action that needs to be taken, like analyzing a website. This is often implied in most tasks.
Detail the Output: Explicitly describe the desired output and any special formatting requirements. Examples include:
- Answer specific questions provided in the prompt.
- Indicate if a keyword is mentioned with a "Yes" or "No."
- List items separated by commas.
- Return numerical ratings or binary responses (e.g., "Yes" or "No").
Avoid Complex Multitasking: Keeping prompts simpler can lead to more consistent and accurate results. Overly complex tasks with multiple parts can lead to misunderstandings and poor performance.
Respect Format Limitations: Be wary that certain format requests, like deriving multiple types of information in a single query, may not be supported. If a specific format is required (only numbers, a certain range, etc.), include such directives in the prompt.
Prompting Examples:
- "Take a company website and answer the following: 'What services does this company offer?'"
- "Return 'Yes' if the website mentions 'sustainable energy', otherwise return 'No'."
By adhering to these guidelines, you can create tasks that are both clear in intention and structured for optimal outcomes, leveraging the full capabilities of the AI-based system.