Hi,
The following best practices should be considered when building prompt templates:
- Make sure that your prompt templates are concise and easy to understand. Avoid using industry jargon or technical terms. Instead, use natural language and conversation design guidelines.
- To give the LLM more contextual information, ask it to role play as a character, such as a sales or support representative. Then define the character’s goal. For example, include language such as, “You are a marketing executive who wants to invite major customers to a live event.”
- Iterate on your prompt templates. Try achieving the same goal using different templates to see how the parts impact the model’s response. Get end-user feedback to see how well your prompt templates generate the desired response.
- Choose a style, and stick to it. When you use a consistent writing style in your prompt templates, the LLM generates consistent responses. Your writing style is shaped by your word choice, intensifiers, emojis, and punctuation.
- To help the LLM differentiate between context and instructions, create an instructions section in your prompt template. On a separate line, enter Instructions:, then surround your instructions with triple quotes (""").
- Include direct instructions for the LLM to only generate the expected type of content. For example, if you want the LLM to draft an email, add instructions such as, “Follow these instructions strictly to generate only the message to be sent to the customer.” These instructions prevent the LLM from generating a response about the process of creating content, instead of just generating the content that you want.
- Start with one of the templates in the Example Prompt Template Library, and customize it to fit your specific needs. Study the language that the templates use, especially the text related to writing style. You can use similar phrasing in your own templates.
Here you can take a look at the example prompt template library
Reference:
