Google

Few-shot learning is facilitated by incorporating multiple demonstrations into a prompt. By adding more examples than the single one used in one-shot learning, the model is given a richer context to understand and execute the desired task.

Enabling Few-Shot Learning with Multiple Demonstrations

Based on the provided scenario, analyze the primary weakness of the current prompt. Describe how you would modify the prompt to improve its classification accuracy for ambiguous tickets and explain the reasoning behind your proposed changes.

Improving a Ticket Classification Prompt

A machine learning engineer is designing a prompt to extract specific financial data (e.g., revenue, net income) from unstructured company earnings reports. They observe that when they provide only one example of a report and its extracted data, the model often misses data points or extracts incorrect information from new reports. Explain the reasoning behind why providing three to five diverse examples, instead of just one, would likely lead to more accurate and reliable data extraction.

Justifying the Use of Multiple Demonstrations

To enable few-shot learning for a grammatical error correction task, a prompt can include multiple demonstrations. For example, after establishing the system's role, the prompt provides several examples of incorrect sentences paired with their corrected versions:

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

**DEMO1** You will be provided with a sentence in English. The task is to output the correct sentence.
Input: There is many reasons to celebrate.
Output: There are many reasons to celebrate.

**DEMO2** You will be provided with a sentence in English. The task is to output the correct sentence.
Input: Me and my friend goes to the gym every day.
Output: My friend and I go to the gym every day.

**USER** 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.
Output:

This structure allows the language model to identify the pattern of correcting grammatical errors from the given examples before predicting the output for the new input.

Example of a Few-Shot Prompt for Grammar Correction

A developer is prompting a large language model to convert customer reviews into a structured summary containing only the product's pros and cons. The initial prompt includes a single example of a review and its corresponding summary, but the model's outputs are inconsistent. Which of the following changes to the prompt is most likely to improve the model's reliability in generating the desired structured summaries?

Learn Before

Related