Google

A prompt template for refining a Large Language Model's output explicitly states the inputs the model will receive to guide the revision. For example, the model can be instructed with the following template: 'You will be provided with a question, its corresponding response, and some feedback on this response.' This structure ensures the LLM uses all three components—the original question, the initial response, and the feedback—to generate an improved answer.

Example of a Prompt Template for Response Refinement

You are designing a system for a Large Language Model (LLM) to improve its own answers. You have the following three pieces of information:

- **Original Question:** 'Explain the process of photosynthesis in simple terms.'
- **Initial Response:** 'Photosynthesis is how plants eat sunlight.'
- **Feedback on Response:** 'The response is too brief and anthropomorphizes plants by saying they 'eat'. It should mention key inputs like water and carbon dioxide, and the output of glucose and oxygen.'

Based on these components, write a complete and effective prompt that instructs the LLM to generate a revised and improved answer.

Constructing a Refinement Prompt

A developer is using a Large Language Model (LLM) to iteratively improve its answers. They provide the LLM with feedback on its initial response using the following prompt: 'Here is some feedback on your last answer. Please provide a better response.' The developer observes that the LLM's revised answers often miss the point of the original user query or seem to ignore the context of its first attempt. Based on the principles of effective prompt design for response refinement, analyze the flaw in the developer's prompt and explain how it leads to the observed issue.

Diagnosing a Flawed Refinement Prompt

A developer is building a system where a Large Language Model (LLM) iteratively improves its own answers. The process involves three stages: 1) The LLM generates an initial response to a user's question. 2) The LLM generates critical feedback on its own initial response. 3) The LLM uses all the information to generate a final, improved response. Which of the following prompt structures, given to the LLM in the third stage, is best designed to produce a high-quality, refined answer?

Learn Before

Related