zhou-2023-ape-instructions-per-dataset

IN premise — summaries/2026/08/24/zhou-2023-context-faithful-prompting-sR-references.md

Created 2026-08-25T02:59:10+00:00

The top-1 APE-generated instructions used were: Natural Questions: 'Read the given information and answer the corresponding question.', Re-TACRED: 'Identify the relationship between two entities from a list of options.', RealTime QA: 'Answer a question based on the provided input-output pairs.'

Summary

The APE process produced simple, task-specific prompts for each dataset — essentially telling the model to answer a question, classify a relationship, or respond based on examples, depending on what the task demands. This matters because it shows the generated instructions are minimal and closely matched to each benchmark's format, meaning the system's performance gains (or lack thereof) can be attributed to the model's reasoning rather than to elaborate prompt engineering.