Conceptpresentation robustness
Presentation Robustness
An evaluation property measuring whether model evaluation results are insensitive to presentation variables unrelated to the evaluation target, such as option order, identifiers, author declarations and framing wording. It constitutes a requirement for out-of-distribution stability.
Timeline (last 90 days)
Nov 5
An evaluator has three independent properties: repeatability (whether results are stable under the same conditions), presentation robustness (whether results are unaffected by irrelevant presentation changes), and human alignment (whether evaluation results are consistent with human preferences), and there is no entailment relationship among the three.
Unverified50%