Actionable Hallucination Detection: Translating Latent Uncertainty into Agentic Critique

📄 arXiv: 2608.10430v1 📥 PDF

作者: Sanidhya Vijayvargiya, Rahul Lokesh

分类: cs.LG, cs.AI

发布日期: 2026-08-11

备注: 22 pages, 6 figures

💡 一句话要点

提出Latent Critic以解决大语言模型的幻觉检测问题

🎯 匹配领域: 支柱一：机器人控制 (Robot Control) 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 大语言模型 幻觉检测 低秩适配器 实时反馈 信号放大 自我修正 AI代理

📋 核心要点

  1. 现有方法在检测大语言模型的幻觉行为时存在局限，无法实时定位和修正幻觉，导致用户体验不佳。
  2. 本文提出的Latent Critic通过与基础LLM并行工作，重构残差流以增强潜在信号，从而实现更精确的幻觉检测。
  3. 实验结果表明，Latent Critic在幻觉定位上达到了0.966的AUROC和超过80%的准确率，显著优于现有检测器。

📝 摘要（中文）

大语言模型（LLMs）作为AI代理在用户规范基础上经常出现失败，执行幻觉化的、不期望的动作。现有检测方法无法提供可操作的实时修正，或无法定位幻觉，或推理延迟过高。本文提出Latent Critic，一个轻量级的低秩适配器（LoRA），与冻结的基础LLM生成并行运行，主动重构变换器的残差流，放大潜在的基础信号，并将其转化为局部的自然语言反馈。通过优化基础模型的固有不确定性信号，这种潜在空间的操作实现了可靠的、细粒度的检测，而无需额外的推理循环。实验结果显示，Latent Critic在Qwen和Llama模型上显著提高了检测和下游任务的效果。

🔬 方法详解

问题定义：本文旨在解决大语言模型在执行用户指令时出现的幻觉行为，现有检测方法无法有效定位和实时修正这些幻觉，影响了模型的可靠性和用户体验。

核心思路：Latent Critic通过与基础LLM的生成过程并行运行，重构变换器的残差流，放大潜在的基础信号，并将其转化为局部的自然语言反馈，从而实现更高效的幻觉检测。

技术框架：该方法的整体架构包括一个轻量级的低秩适配器（LoRA），与冻结的基础LLM并行工作，主要模块包括信号放大、反馈生成和实时检测。

关键创新：Latent Critic的最大创新在于其能够在不增加额外推理延迟的情况下，实时重构潜在空间中的不确定性几何，使得幻觉检测更加可靠和细粒度。

关键设计：在设计中，Latent Critic采用了低秩适配器的结构，优化了信号放大过程，并通过激活补丁和层级探测进行机制分析，确保了其在不同模型上的有效性。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示，Latent Critic在幻觉检测中表现优异，达到了0.966的AUROC和超过80%的定位准确率，显著优于同规模的外部检测器和被动内部探测器，展示了其在实时应用中的优势。

🎯 应用场景

该研究的潜在应用领域包括智能助手、自动化客服和其他依赖于大语言模型的AI系统。通过提高幻觉检测的准确性和实时性，Latent Critic能够显著提升用户体验，减少不期望的行为发生，未来可能推动更安全和可靠的AI代理的发展。

📄 摘要（原文）

Large Language Models (LLMs) deployed as AI agents frequently exhibit user specification-grounding failures, executing hallucinated, undesired actions to force a resolution rather than expressing uncertainty. Existing detection methods fail to provide actionable, real-time correction as they either do not localize the hallucinations, or incur prohibitive inference latency. We introduce the Latent Critic, a lightweight low-rank adapter (LoRA) that operates concurrently with a frozen base LLM's generation to actively restructure the transformer's residual stream---amplifying latent grounding signals and translating them into localized, natural language feedback within a single sequence. By refining the base model's native uncertainty signals, this manipulation of the latent space enables reliable, granular detection without the overhead of secondary inference loops. Mechanistic analysis via activation patching and layer-wise probing shows that this rank-invariant behavior restructures pre-existing uncertainty geometry into a linearly separable representation that transfers more reliably than base model representations alone. Using tool-calling as an instantiation of granular hallucinations, we validate the detection and downstream improvements enabled by the Latent Critic architecture across Qwen and Llama-based models. Demonstrating superior real-time efficacy, our approach significantly outperforms equivalent-scale fine-tuned external detectors, semantic entropy baselines, and passive internal probes in isolating hallucinations, achieving 0.966 AUROC and >80% accuracy in localization (e.g., ungrounded: date). When deployed in a closed-loop ReAct environment, the Critic acts as a negligible latency guardrail, intercepting hallucinations before execution to prevent undesired actions while simultaneously leveraging this specific localized feedback to enable efficient agent self-correction.