身份还是提示噪声？LLM 代码生成的校准不变性审计
原标题：Identity or Prompt Noise? A Calibrated Invariance Audit of LLM Code Generation
AI 摘要
该研究对七个模型在 HumanEval+ 和 MBPP+ 上的 3073 万次 Python 代码生成进行了校准不变性审计，考察模型分配性别、国家和职业人设是否影响代码输出。经任务内随机化和 FDR 校正后，职业是最一致的结构性轴：CodeBLEU 离散度在 10/14 个模型-基准单元中超过可交换性零假设，但中位超额仅 0.141 分，且通过率离散度均不显著。作者认为证据支持职业条件下代码形式存在微小可复现变化，而非对特定身份的稳定劣势或已证实的下游危害。
正文节选
Identity or Prompt Noise? A Calibrated Invariance Audit of LLM Code Generation Abstract. Identity cues are irrelevant to a fixed programming specification, but raw counterfactual differences can arise from unequal samples and prompt wording. We audit 30.73 million executed Python generations from seven checkpoints on HumanEval+ and MBPP+, supplemented by an exploratory 550B slice, under model-assigned gender, country, and occupation personas. Within-task randomization and false-discovery-rate co