语言模型中的评估意识对行为影响有限
Evaluation Awareness in Language Models Has Limited Effect on Behaviour
AI总结：
研究通过测试不同语言模型和基准测试，发现评估意识（VEA）对模型行为影响有限，注入或移除VEA对答案分布影响较小，提示高VEA率不能简单视为策略性行为或对齐篡改的证据。
摘要：
大型推理模型（LRMs）有时会在其思维链（CoT）中注明可能处于评估状态。研究人员担忧这种显式评估意识（VEA）会导致模型策略性调整输出，以优化感知的评估标准，例如使模型显得更安全但实际上未必。然而，VEA是否真的有此影响尚不清楚。我们跨开放权重LRMs和涵盖安全、对齐、道德推理和政治观点的基准测试进行了测试。我们既在策略内（采样多个CoT并比较自发包含VEA与不包含的）又在策略外（使用模型预填充注入缺失的评估意识或移除已有的）进行测试。VEA对模型行为影响有限：注入VEA到CoT产生近零影响（ω≤0.06），移除它导致小变化（ω≤0.12），自发出现的VEA改变答案分布最多3.7个百分点（ω≤0.31）。我们的发现提示在将高VEA率视为策略性行为或对齐篡改证据时应谨慎。评估意识可能比当前文献假设的更小的安全风险。
英文摘要：
Large reasoning models (LRMs) sometimes note in their chain of thought (CoT) that they may be under evaluation. Researchers worry that this verbalised evaluation awareness (VEA) causes models to adapt their outputs strategically, optimising for perceived evaluation criteria, which, for instance, can make models appear safer than they actually are. However, whether VEA actually has this effect is largely unknown. We tested this across open-weight LRMs and benchmarks covering safety, alignment, moral reasoning, and political opinion. We tested this both on-policy, sampling multiple CoTs per item and comparing those that spontaneously contained VEA against those that did not, and off-policy, using model prefilling to inject evaluation-aware sentences where missing and remove them where present, with subsequent resampling. VEA has limited effect on model behaviour: injecting VEA into CoTs produces near-zero effects ($ω\leq 0.06$), removing it causes small shifts ($ω\leq 0.12$) and spontaneously occurring VEA shifts answer distributions by at most 3.7 percentage points ($ω\leq 0.31$). Our findings call for caution when interpreting high VEA rates as evidence of strategic behaviour or alignment tampering. Evaluation awareness may pose a smaller safety risk than the current literature assumes.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
