Google

Constrained prompts direct a model to perform a task while adhering to specific limitations. For instance, a prompt for text summarization could impose a length constraint, such as 'Summarize the following article in no more than 50 words,' or a structural constraint, like 'Summarize this text in three sentences.'

Examples of Constrained Prompts for Text Summarization

Analyze the effectiveness of the initial instruction provided in the case study below and then rewrite it to better achieve the desired outcome.

Improving a Prompt for a Specific Use Case

A project manager needs to create a high-level overview of a detailed technical report for a non-technical executive audience. The goal is a summary that is brief, easy to understand, and highlights the main business implications.

Consider two possible instructions to provide to a language model:

*   **Instruction A:** "Summarize the following technical report."
*   **Instruction B:** "Summarize the following technical report in three simple sentences, avoiding technical jargon. Focus on the key findings and their business implications."

Which instruction is more likely to generate a summary that meets the project manager's needs? Justify your choice by analyzing the specific components of each instruction and explaining why one is more effective than the other in this context.

Evaluating Prompt Effectiveness for Summarization

An example of a constrained text transformation task involves prompting a language model with the instruction 'Summarize the following article in no more than 50 words:', followed by a source text detailing urban sustainability challenges. The model is expected to output a concise overview, such as 'Urban areas are grappling with sustainability challenges, such as rising pollution, traffic congestion, and infrastructure demands...', adhering strictly to the prescribed length constraint.

Example of an Article Summarization Prompt with a Word Constraint

A user wants to generate a summary of a long business report that consists of exactly three bullet points. Which of the following instructions is best designed to guide a language model to produce this specific output?

Learn Before

Related