Google

When using a Large Language Model to generate an initial set of prompts, two main strategies can be employed. The first approach is to provide the LLM with a human-written description of the task, which is a direct method but is contingent on having such a description. An alternative strategy is to provide the LLM with input-output examples of the task, which allows the model to infer the prompt's purpose without needing a formal task description.

Comparison of LLM-based Prompt Generation Methods for Initialization

Given the team's situation described below, which of the two primary methods for using a large language model to generate an initial set of prompts would you recommend: providing the model with a human-written task description or providing it with the input-output examples? Justify your recommendation by evaluating the suitability of each method in this specific context.

Choosing an Initial Prompt Generation Strategy

A developer is using a large language model to generate an initial set of prompts for a new task: classifying customer support emails as 'Urgent' or 'Not Urgent'. The developer provides the model with a detailed, formal definition of what constitutes an 'Urgent' email, including criteria like keywords, response time expectations, and escalation protocols. The resulting prompts are logically correct but fail to capture the subtle nuances and varied language used by real customers, leading to poor performance. Analyze the potential weakness of the chosen initialization strategy in this context and propose a more effective alternative strategy, explaining why it would be better suited for this task.

Diagnosing Prompt Initialization Failure

A machine learning engineer is tasked with creating a set of initial prompts for a novel text summarization task. The goal is to summarize complex legal documents into plain language, a process for which a precise, formal description is challenging to write. However, the engineer has access to a large, curated dataset of several hundred legal documents and their corresponding expert-written plain-language summaries. Which initialization strategy would be most effective in this situation, and why

Learn Before

Related