zhou-2023-gpt35-mr-reduction-zero-shot

IN premise — summaries/2026/08/24/zhou-2023-context-faithful-prompting-s4-experiments.md

Created 2026-08-25T02:59:11+00:00

For GPT-3.5 (text-davinci-003), the zero-shot to OPIN+INSTR+Counterfactual transition yielded a 32.2% reduction in MR for MRC and a 10.9% reduction for RE.

Summary

Switching GPT-3.5 from plain zero-shot prompting to a structured format that combines opinion, instruction, and counterfactual framing cuts its error rate by roughly a third on reading comprehension tasks and by about 11% on relation extraction. This is a direct empirical data point showing that prompt structure is a meaningful lever for reducing errors in this model, with the effect being much stronger on comprehension-style tasks than on extraction-style ones.