Custom-built GPTs can be powerful tools for enhancing any workflow, but they necessitate well-crafted instructions to perform as desired. In this post, we walk through six strategies, each with practical tips, designed to help you utilize these models more effectively; sourced from OpenAI's excellent piece on prompt engineering.

1. Write Clear Instructions

The precision of your instructions dramatically impacts the quality of the model's responses.

Tip: Include Details in Your Query

Examples

• ‍‍Worse: "How do I add numbers in Excel?"
• Better: "How do I add up a row of dollar amounts in Excel and automatically display the totals in a column named 'Total'?"

Tip: Ask the Model to Adopt a Persona

Example

• ‍‍"Respond with a playful tone, incorporating humor in each paragraph."

Tip: Use Delimiters for Distinct Parts

Example:‍

• "Summarize the text within triple quotes in a haiku format."

Tip: Specify Steps for Task Completion

Example:‍

• "Step 1: Summarize the text in one sentence. Step 2: Translate that summary into Spanish."

Tip: Provide Examples

Example:‍

• "Demonstrate how to respond to inquiries in a manner similar to this specific example."

Tip: Specify Desired Output Length

Example:‍

• "Summarize the text in approximately 50 words."

2. Provide Reference Text

Combat the model's tendency to fabricate answers by supplying reference material.

Tip: Answer Using Reference Text

Example:‍

• "Use the provided articles to answer questions. If unsure, state 'I could not find an answer.'"

Answer with Citations from Reference Text

Example:‍

• "Include citations from the provided document to support your answers."

3. Split Complex Tasks into Simpler Subtasks

Breaking down tasks can reduce error rates and enhance clarity.

Tip: Use Intent Classification

Example:‍

• "Classify customer service queries into categories like 'Billing' or 'Technical Support', then provide targeted responses."

Summarize Long Documents Piecewise

Example:‍

• "Summarize each chapter separately, then compile these into a comprehensive summary."

4. Give the Model Time to "Think"

Encouraging the model to deliberate can lead to more accurate answers.

Tip: Work Out Solutions Before Concluding

Example:‍

• "First, compute your own solution to the problem, then compare it with the student's solution."

5. Use External Tools

Upload any documents in the knowledge base or use the added capabilites by enabling we-browsing, Dall-E, code interpreter.

Tip: Upload Documents of Best Practices or Your Company Specifications

Example:‍

• "This is an example of a service quote I created for one of my clients, I want you to write it in a similar format for all my other clients."

6. Test Changes Systematically

Consistent testing ensures that modifications truly enhance performance.

Tip: Evaluate with Gold-Standard Answers

Example:‍

• "Compare the model's answers against a set of known facts and evaluate the level of accuracy."

In summary, these strategies and tactics offer a structured approach to interacting with GPT models. By refining the way we pose questions and structure tasks, we can harness the full potential of these powerful tools.