Consider the following instruction given to a language model for a grammar correction task: 'Please fix the grammar.' Based on the principles of creating clear and direct instructions, identify one major weakness of this instruction and rewrite it to be more effective.

Google

A direct instruction for a grammatical error correction task can be formulated as: "You will be provided with a sentence in English. The task is to output the correct sentence." This statement explicitly tells the model what kind of input to expect and what its objective is.

Example of a Task Instruction for Grammatical Error Correction

A user wants to use a language model to correct grammatical errors in their sentences. To get the best and most consistent results, which of the following instructions is the most effective to provide to the model?

A prompt for a grammar correction task can be structured to guide a Large Language Model by defining its persona, the nature of the input, and the expected output. For instance, a complete prompt might look like this:

`You are a helpful assistant, and are great at grammar correction.`

`You will be provided with a sentence in English. The task is to output the correct sentence.`

`Input: She don’t like going to the park.`

This example first establishes the model's role as a grammar expert, then specifies the task, and finally provides the actual input sentence, clearly labeled with the 'Input:' field.

Example of a Persona-based Prompt for Grammar Correction

Analyze the following task instruction intended for a language model. Identify two specific weaknesses in this instruction and explain why each weakness could lead to unpredictable or undesirable results.

Learn Before

Related