zhou-faithfulness-distinct-from-accuracy

IN premise — summaries/2026/08/24/zhou-2023-context-faithful-prompting-s2-related-work.md

Created 2026-08-25T02:59:09+00:00

Faithfulness is defined as the validity of the extraction process (grounding in provided context), distinct from answer accuracy, since a model can be accurate via parametric recall yet unfaithful.

Summary

Just because a model gives the right answer doesn't mean it actually used the evidence you gave it; it might have pulled the answer from its own training data. This distinction matters because a system that looks accurate on the surface can still be ignoring the source material entirely, which breaks trust in any pipeline that depends on citations or context-grounded reasoning.