Bias scores in language-model evaluations often conflate what a model prefers with whether it answers at all. We identify this confound in \textsc{BBQ}, whos…
机构:UIUC
来源:arXiv 2610.17641 | AI4Papers 论文推荐平台