距离崩溃仅一步之遥：指令调制帮助性的脆弱性
One Token Away from Collapse: The Fragility of Instruction-Tuned Helpfulness
AI总结：
研究探讨了指令调制大语言模型在简单约束下的帮助性稳定性，发现其在词汇约束下响应质量显著下降，揭示了指令调制对表面模板的依赖性。
摘要：
指令调制的大语言模型能够生成有帮助的结构化响应，但这种帮助性在简单的词汇约束下是否具有鲁棒性？我们发现简单的词汇约束（禁止单个标点符号或常见词汇）会导致指令调制LLM响应崩溃，失去14-48%的全面性，跨七个模型（7B-70B，开源和闭源模型）表现各异。盲评人类评估显示真实内容损失，信息标准下降1.5-2.3倍，自动化配对比较验证了这一发现。诊断分析表明这是规划失败：双步生成恢复59-96%的响应长度，线性探针预测响应长度（R²=0.51-0.94）在生成前表现显著，而基础模型则无此表现，证明指令调制引入了崩溃的基础结构。基础模型在相同约束下无系统退化，表明指令调制将任务能力与狭窄的表面模板耦合。这种影响延伸到现实部署约束（前缀抑制、企业语气指南、法律合规缓冲、可访问性要求）导致相似退化（-22%至-34%），仅抑制对话开场白（“Certainly!”）导致最脆弱模型40%崩溃，尽管仅限制了开头token。我们进一步发现标准独立LLM作为判断者评估仅检测到3.5%的质降，而配对评估显示23%，揭示了当前评估实践中的方法盲点。
英文摘要：
Instruction-tuned large language models produce helpful, structured responses, but how robust is this helpfulness under trivial constraints? We show that simple lexical constraints (banning a single punctuation character or common word) cause instruction-tuned LLMs to collapse their responses, losing 14--48\% of comprehensiveness across seven models spanning five families (7B--70B, open- and closed-weight). A blinded human evaluation with 10 STEM-trained evaluators confirms genuine content loss, with information criteria degrading $1.5$--$2.3\times$ more than surface criteria, a finding corroborated by over 4,100 automated pairwise comparisons (77--100\% baseline preference) across three LLM judges from two model families. Diagnostic analysis identifies this as a \emph{planning failure}: two-pass generation recovers 59--96\% of response length, and linear probes on prompt representations predict response length with $R^2 = 0.51$--$0.94$ before generation begins. The same probes yield negative $R^2$ on base models, confirming that instruction tuning introduces the representational structure underlying the collapse. Base models show no systematic degradation under identical constraints, demonstrating that instruction tuning couples task competence to narrow surface-form templates. The effect extends to realistic deployment constraints (preamble suppression, corporate tone guidelines, legal compliance hedging, accessibility requirements) causing comparable degradation ($-$22\% to $-$34\%), with suppressing the conversational opener alone (``Certainly!'') causing 40\% collapse on our most fragile model despite restricting only the opening tokens. We further show that standard independent LLM-as-judge evaluation detects only a 3.5\% quality drop where pairwise evaluation reveals 23\%, exposing a methodological blind spot in current evaluation practice.
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。
