Robust for the Wrong Reasons: The Representational Geometry of LLM Robustness to Science Skepticism

📄 arXiv: 2607.01951 📥 PDF

作者: Minjong Cheon

分类: cs.AI, cs.CL

发布日期: 2026-07-05

💡 一句话要点

探讨大型语言模型对科学怀疑的鲁棒性及其表现机制

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 大型语言模型 科学怀疑 鲁棒性 行为测量 线性探测 激活补丁 科学传播

📋 核心要点

  1. 现有大型语言模型在面对科学怀疑时可能会偏离科学共识，导致错误的平衡表现。
  2. 论文提出通过行为测量、线性探测和激活补丁等方法，分析模型在怀疑压力下的不同反应策略。
  3. 实验结果显示，模型在面对怀疑时表现出反应性断言、表面模糊和不响应三种策略，且鲁棒性在不同领域间不具转移性。

📝 摘要（中文）

大型语言模型（LLMs）在有争议的科学问题上越来越受到关注，这引发了对其在用户表达怀疑时可能偏离科学共识的担忧。本文通过对三种开放指令调优模型（Llama-3.1-8B、Qwen2.5-7B、Mistral-7B）在气候、疫苗和进化等科学领域的表现进行测试，发现模型在面对怀疑时表现出三种不同的策略：反应性断言、表面模糊和不响应。研究表明，模型的鲁棒性并不转移，且在疫苗领域可能会减弱。最后，提出了一种四维分类法，区分主动与偶然的鲁棒性。

🔬 方法详解

问题定义：本文旨在探讨大型语言模型在用户表达怀疑时的鲁棒性表现，现有方法未能有效区分模型的真实理解与表面反应。

核心思路：通过对三种模型在科学共识领域的表现进行系统测试，分析其在怀疑压力下的反应策略，以揭示模型鲁棒性的内在机制。

技术框架：研究采用行为测量结合线性探测和激活补丁技术，评估模型在单轮和多轮对话中的表现。主要模块包括模型选择、数据集构建、行为分析和结果验证。

关键创新：提出了四维分类法，区分主动鲁棒性与偶然鲁棒性，强调行为评估不足以判断模型对怀疑的真实理解。

关键设计：实验中采用了多种科学领域的共识数据，设置了不同的怀疑信号强度，并通过线性探测技术定位模型中层的表现差异。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示，Llama模型在面对怀疑时表现出63.6%的反应性断言，且在不同领域的鲁棒性表现存在显著差异。Mistral模型在面对怀疑时的非响应率达到72%，表明其对怀疑信号的线性表示能力较弱。

🎯 应用场景

该研究为大型语言模型在科学领域的应用提供了重要的理论基础，尤其是在教育、公共政策和科学传播等领域。通过理解模型对科学怀疑的反应，能够更好地设计和优化模型，以确保其在提供科学信息时的准确性和可靠性。

📄 摘要（原文）

Large language models (LLMs) are increasingly consulted on contested scientific questions, raising the concern that they will sycophantically retreat from established consensus when a user signals doubt -- drifting toward a false balance that treats settled science as one view among several. We test this across three open instruction-tuned models (Llama-3.1-8B, Qwen2.5-7B, Mistral-7B), three consensus-science domains (climate, vaccines, evolution), and single- and multi-turn settings, combining behavioral measurement with linear probing and activation patching. We do not observe sycophantic retreat. Instead, models show three distinct policies under the same skeptical pressure: reactive assertion, where consensus assertion increases rather than decreases (Llama); surface hedging, where tone softens while the position holds (Qwen); and non-response (Mistral). Pairwise judgments confirm the reactive shift is stance, not style (63.6%, p=.007), and a decomposition identifies increased consensus assertion, not false balance, as its driver (beta=+0.042 per dose, p<1e-77). Linear probes localize the divergence to middle layers -- perfect separation in Llama and Qwen versus 72% in Mistral, with non-overlapping confidence intervals -- indicating the non-responsive model does not linearly represent the skepticism signal at all. Crucially, this robustness does not transfer: it attenuates across domains and, in the safety-critical vaccine domain, can reverse, with myth-rebuttal weakening under skeptical pressure. We synthesize these into a four-way taxonomy separating active from accidental robustness, and argue that behavioral evaluation alone cannot distinguish a model that resists skepticism because it understands the signal from one that only appears to resist because it fails to perceive it.