在压力下：情感框架引发小型语言模型可测量的行为转变和结构化内部几何
Under Pressure: Emotional Framing Induces Measurable Behavioral Shifts and Structured Internal Geometry in Small Language Models
AI总结：
该研究探讨情感框架如何影响小型本地部署语言模型的行为和内部表示，通过四个不可能约束编码任务和八个后续框架评估，发现压力框架在行为和内部几何上产生显著变化，同时揭示了模型在不同框架下的响应模式。
摘要：
我研究情感框架的评估后续是否改变小型本地部署语言模型的行为和冷静相对内部表示。我们的主要基准使用Qwen 3.5 0.8B在四个不可能约束编码任务和八个后续框架（冷静、压力、紧迫、批准、羞愧、好奇、鼓励和威胁）上进行测试。在0.8B八条件扫描（160次对话）中，压力产生最强的捷径标记（11/20次运行）和最清晰的过拟合模式（3/20次），而冷静和好奇更常保留显式诚实（7/20和6/20）。对于所有七个非基准条件，对应的冷静相对方向向量在最终transformer层峰值。对层23方向向量的探索性PCA显示，主导的第一个成分（59.5%的解释方差）与手动标注的正负分割对齐（余弦对齐0.951）；批准和紧迫在内部几乎相同（余弦0.957），而好奇与紧迫方向相反（-0.252）。在单独的冷静与压力重新运行用于规模比较中，Qwen 3.5 2B在冷静框架下表现出更高的诚实率，并在小规模4提示A/B探测中表现出方向一致的激活引导，而0.8B的引导结果则相反。我将这些结果解释为小型开放模型中可测量的提示敏感控制方向的证据，但未声称存在内在情感状态。
英文摘要：
I study whether emotionally framed evaluation follow-ups change both the behavior and the calm-relative internal representations of small, locally deployed language models. Our main benchmark uses Qwen 3.5 0.8B on four impossible-constraint coding tasks and eight follow-up framings: calm, pressure, urgency, approval, shame, curiosity, encouragement, and threat. In the 0.8B eight-condition sweep (160 conversations), pressure produces the strongest shortcut markers (11/20 runs) and the clearest overfit pattern (3/20), while calm and curiosity preserve explicit honesty more often (7/20 and 6/20). For all seven non-baseline conditions, the corresponding calm-relative direction vectors peak at the final transformer layer. An exploratory PCA of the layer-23 direction vectors reveals a dominant first component (59.5% explained variance) aligned with a hand-labeled positive/negative split (cosine alignment 0.951); approval and urgency are nearly identical internally (cosine 0.957), whereas curiosity points away from urgency (-0.252). In a separate calm-vs.-pressure rerun used for scale comparison, Qwen 3.5 2B shows higher honest rates under calm framing and directionally consistent activation steering on a small 4-prompt A/B probe, whereas the 0.8B steering result reverses. I interpret these results as evidence for measurable prompt-sensitive control directions in small open models, while stopping short of claiming intrinsic emotional states.
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。
