1
The Wording Effect: Quantifying Two-Way Drift in LLM Benchmark Performance
同一个问题换个问法，模型答案就可能翻车？量化措辞漂移，揭示基准测试的隐藏盲区。
arXiv:2608.11694v1 Announce Type: cross Abstract: A benchmark score comes from a single phrasing of each problem. That single phrasing is treated as i…