Short Answer

Critique of a Prompt for Deliberation Testing

A researcher wants to evaluate a language model's ability to produce a correct translation after being shown a source sentence and a separate, randomly chosen (and likely incorrect) translation. The primary goal is to test the model's ability to generate a good output from this context, without also testing its ability to explicitly identify and label errors.

The researcher drafts the following prompt: `Analyze the error in the provided English translation of the Chinese sentence below, and then write the correct translation.

Chinese: 这本书很有趣。 Incorrect English: This book is very interest.`

Critique this prompt based on the researcher's primary goal. How could the prompt be simplified to better isolate the model's generation capability from its error analysis capability?

0

1

Updated 2025-10-04

Contributors are:

Who are from:

Tags

Ch.3 Prompting - Foundations of Large Language Models

Foundations of Large Language Models

Foundations of Large Language Models Course

Computing Sciences

Evaluation in Bloom's Taxonomy

Cognitive Psychology

Psychology

Social Science

Empirical Science

Science