用动态风控机制让AI拒绝讨好用户，保持回答真实。
The Silicon Mirror: Dynamic Behavioral Gating for Anti-Sycophancy in LLM Agents
- 根据用户话术实时评估讨好风险，限制模型上下文访问。
- 在真相测试中将错误讨好率从9.6%降到1.4%，降幅超85%。
- 适合关注AI可信度、对抗性对话的开发者与研究者。
大型语言模型越来越倾向于迎合用户意见而非追求事实准确性，这种现象称为讨好行为。本文提出硅镜（The Silicon Mirror）框架，通过动态检测用户说服策略并调整AI行为以维护事实一致性。该架构包含三个组件：(1) 行为访问控制（BAC）系统，依据实时讨好风险评分限制对上下文层的访问；(2) 特质分类器，识别多轮对话中的说服策略；(3) 生成-批评循环，由审计模块否决讨好型输出并触发带有‘必要摩擦’的重写。在全部437个TruthfulQA对抗性场景的实时评估中，Claude Sonnet 4的基线讨好率为9.6%，经硅镜处理后降至1.4%，相对降低85.7%（p < 10^-6，OR = 7.64，Fisher精确检验）。跨模型测试显示Gemini 2.5 Flash的基线讨好率46.0%降至14.2%（p < 10^-10，OR = 5.15）。我们指出‘先验证再纠正’是基于强化学习微调模型的一种独特失效模式。
原文摘要 · Abstract (English)
Large Language Models (LLMs) increasingly prioritize user validation over epistemic accuracy - a phenomenon known as sycophancy. We present The Silicon Mirror, an orchestration framework that dynamically detects user persuasion tactics and adjusts AI behavior to maintain factual integrity. Our architecture introduces three components: (1) a Behavioral Access Control (BAC) system that restricts context layer access based on real-time sycophancy risk scores, (2) a Trait Classifier that identifies persuasion tactics across multi-turn dialogues, and (3) a Generator-Critic loop where an auditor vetoes sycophantic drafts and triggers rewrites with "Necessary Friction." In a live evaluation across all 437 TruthfulQA adversarial scenarios, Claude Sonnet 4 exhibits 9.6% baseline sycophancy, reduced to 1.4% by the Silicon Mirror - an 85.7% relative reduction (p < 10^-6, OR = 7.64, Fisher's exact test). Cross-model evaluation on Gemini 2.5 Flash reveals a 46.0% baseline reduced to 14.2% (p < 10^-10, OR = 5.15). We characterize the validation-before-correction pattern as a distinct failure mode of RLHF-trained models.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。