Tag
This paper audits the reproducibility of LLM evaluation conclusions, finding that small-sample evaluations are unstable and can mislead, and it recommends reporting additional metrics for reliability.