OpenAI just dropped a guide on how to make ChatGPT and LLM perform better, and we’ve got the inside scoop. Here are the tricks straight from the source:
Be Crystal Clear with Instructions
- Details matter. Make your questions specific for meaningful answers.
- Give the chatbot a role. For example, say, “You are an expert…”
- Use delimiters (triple quotes) to pinpoint specific parts of your question.
- Spell out the steps needed to finish a task.
- Provide examples.
- Specify how long you want the answer to be.
Reference Texts for Accuracy
Models can goof up on tricky questions or specific info like quotes and URLs. Giving them reference texts can cut down on mistakes.
- Guide the model to respond based on a reference text.
- Ask the model to reply with quotes from a reference text.
Break Down Bigger Tasks
Complex tasks can trip up the model. Break them into smaller, more manageable parts.
- Identify the most relevant instructions for a user’s request by selecting, questioning, and clarifying.
- Summarize long documents in parts and build a complete summary step by step.
Let the Model Take Its Time
Models mess up more when they rush. Give them a moment to think.
- Ask the model to generate its solution before checking an existing one.
- Use an internal monologue or series of questions to hide the model’s thinking.
- Check if the model missed anything in previous rounds.
SYSTEM
Follow these steps to answer the user queries.Step 1 – First work out your own solution to the problem. Don’t rely on the student’s solution since it may be incorrect. Enclose all your work for this step within triple quotes (“””).
Step 2 – Compare your solution to the student’s solution and evaluate if the student’s solution is correct or not. Enclose all your work for this step within triple quotes (“””).
Step 3 – If the student made a mistake, determine what hint you could give the student without giving away the answer. Enclose all your work for this step within triple quotes (“””).
Step 4 – If the student made a mistake, provide the hint from the previous step to the student (outside of triple quotes). Instead of writing “Step 4 – …” write “Hint:”.
USER
Problem Statement: <insert problem statement>Student Solution: <insert student solution>
Example prompt from OpenAI for an inner monologue
OpenAI doesn’t talk about some peculiar prompt additions like saying “Take a deep breath” or emotionally pressuring the chatbot in this guide.
External Tools
To overcome the usual limitations of large language models, you can level up by employing additional tools like text search systems or code execution programs. These hybrid models, combining language skills with external tools, have the potential to be way more potent than standalone language models. Here’s the game plan for this strategy:
- Smart Knowledge Retrieval with Embedded Search:
- Utilize embedded search systems to efficiently fetch relevant information.
- Precision Boost with Code Execution:
- Leverage code execution to perform precise calculations or interact with external APIs for added functionality.
- Unlocking Special Powers with Model Access:
- Grant the model access to specific functions, enhancing its capabilities and making it a dynamic force.
Moreover, OpenAI suggests a more systematic approach to assess the effectiveness of frequently used prompts. Rather than relying on intuitive feelings, consider these structured evaluations:
- Reflect Actual Usage:
- Evaluate prompts based on how they are actually used in real-world scenarios.
- Include Diverse Test Cases:
- Encompass a variety of test cases to ensure the robustness and adaptability of your prompts.
- Automation and Repetition:
- Design evaluations that are easy to automate or repeat, allowing for consistent and reliable assessments over time.
By adopting this methodical evaluation process, you can gain a clearer understanding of how well your prompts perform across different situations and ensure their effectiveness in a range of contexts.
Conclusion
OpenAI has shared valuable tips for using AI language models, specifically ChatGPT, more effectively. They’ve outlined six strategies to enhance performance and obtain more accurate results from these language models. By following these strategies, users can optimize their interactions with AI models, leading to improved outcomes and a better overall experience.
