mquad-t-extended-golden-labels-text-davinci-003

IN premise — summaries/2026/08/24/zhong-2023-mquake-sR-references-chunk-2.md

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

MQuAKE-T uses extended golden labels generated via GPT-3.5 (text-davinci-003) that broaden the acceptable answer set to include outdated but historically valid answers for smaller models.

Summary

The MQuAKE-T benchmark grades answers with a broader set of accepted responses, adding answers that were factually correct in the past even if they are no longer current, because smaller models often lack access to the very latest information. This makes the evaluation fairer across different model sizes by not penalizing a model for knowing slightly older facts.