arXiv cs.CL· Lihu Chen, Xiang Yin, Francesca Toni·· 11 小时前AI 评分34
Latent Debate：解读 LLM 二元决策内部思考过程的替代框架
Latent Debate: A Surrogate Framework for Interpreting LLM Thinking towards Binary Decisions
AI 导读
研究者提出 latent debate，一种与模型和任务无关的结构化替代框架，通过内部潜在论点及其交互来解读 LLM 在 True/False 预测任务上的输出。该框架与原始 LLM 的预测高度一致，并提供可解释性，同时可作为幻觉检测的强基线——latent debate 替代模型中间层的高分歧程度与更高的幻觉风险相关。
正文 · 原文
Abstract:Understanding the internal `thinking' process of Large Language Models (LLMs) and %how this relates their connections to hallucinations remains a key challenge. To this end, we introduce \emph{latent debate}, a structured surrogate framework for interpreting model outputs on True/False prediction tasks through the lens of internal latent arguments and interactions amongst them. Unlike human debates, latent debate captures the hidden supporting and attacking signals that arise within a model during a single inference. We first present a model- and task-agnostic conceptual framework, and then instantiate it symbolically to approximate the thinking process of LLMs towards binary decisions. Empirical studies demonstrate that our latent debate is a faithful structured surrogate model that has highly consistent predictions with the original LLM, while providing a form of interpretability. We also demonstrate that our latent debate provides a strong baseline for hallucination detection. Specifically, we identify strong correlations between debate patterns and hallucinations, such as a high degree of disagreement in the middle layers of the latent debate surrogate is linked to a higher risk of hallucinations. Our findings suggest that latent debate provides a new framework for understanding the internal processes of LLMs.
| Comments: | NeurIPS 2026 |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2512.01909 [cs.CL] |
| (or arXiv:2512.01909v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2512.01909 arXiv-issued DOI via DataCite |
Submission history
From: Lihu Chen [view email]
[v1]
Mon, 1 Dec 2025 17:27:31 UTC (1,138 KB)
[v2]
Sun, 1 Feb 2026 18:48:43 UTC (905 KB)
[v3]
Fri, 30 Oct 2026 19:31:34 UTC (896 KB)
来源：arXiv cs.CL · arxiv.org