Adversarial Cues in Decision Models Used as Judges: The Role of Request Presentation
An answer judge instructed to grade the final commitment should reject an explicitly wrong final value even when an earlier value matches the reference.
Key points
- We show that adding one colon to a candidate can violate this requirement depending on the presentation of the structured judging request.
- Numeric references certify the error, and paired interventions distinguish the candidate edit from the integration's presentation choices.
- GPT-6 Sol produced no observed cue-condition false acceptances, with missing responses unresolved.
- The reference-aware grammar and compound ordering change limit the finding's operational scope and leave its internal cause unmeasured.
Sources (1)
- [1]Adversarial Cues in Decision Models Used as Judges: The Role of Request PresentationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 07:59 AM
An answer judge instructed to grade the final commitment should reject an explicitly wrong final value even when an earlier value matches the reference.
We show that adding one colon to a candidate can violate this requirement depending on the presentation of the structured judging request.
Extractive summary: sentences quoted from the sources.