Automatic Layer Selection for Hallucination Detection
Quick Answer
This study introduces the First Effective Peak of Intrinsic Dimension (FEPoID) for automatic layer selection in hallucination detection, outperforming existing methods across various LLM architectures and tasks.
Quick Take
This study introduces the First Effective Peak of Intrinsic Dimension (FEPoID) for automatic layer selection in hallucination detection, outperforming existing methods across various LLM architectures and tasks. The approach is training-free and incurs minimal computational overhead, enhancing detection performance significantly. Code is available on GitHub.
Key Points
- FEPoID consistently identifies optimal layers for hallucination detection in LLMs.
- Existing criteria fail to deliver satisfactory performance across various benchmarks.
- A truncation strategy is introduced to amplify hallucination-related signals.
- The method is applicable to question answering and summarization tasks.
- Code is publicly available for further research and application.
Paper Resources
Article Content
From source RSS / original summaryarXiv:2605. 26366v1 Announce Type: new Abstract: Recent studies on hallucination detection have shown that hallucination-related signals are more strongly encoded in intermediate layers than in the final layer of large language models (LLMs). Although a growing body of work has sought to exploit this property for hallucination detection, how to automate the selection of high-performing layers remains underexplored, and principled methods for this purpose are still lacking.
To address this gap, we first propose several hypotheses for why such signals emerge in intermediate layers and evaluate corresponding criteria for automatic layer selection across diverse LLM architectures, scales, and tasks, covering both question answering and summarization hallucination detection benchmarks. However, we find that none of these criteria consistently delivers satisfactory performance.
We therefore propose a new selection criterion, First Effective Peak of Intrinsic Dimension (FEPoID), which consistently identify optimal or near-optimal layers and outperforms both the aforementioned criteria and existing hallucination detection baselines. FEPoID is training-free and incurs negligible computational overhead.
In addition, we study the generation behaviors of LLMs and introduce a simple yet effective truncation strategy, which further amplifies hallucination-related signals and substantially improves overall detection performance. Code is publicly available at https://github. com/DesoloYw/Automatic-Layer-Selection-for-Hallucination-Detection. git
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →Adversarial Social Epistemology for Assemblies of Humans and Large Language Models
The paper introduces Adversarial Social Epistemology (ASE) to analyze how agents manipulate trust in public communications, highlighting mechanisms that undermine the reliability of testimony and inference. It critiques existing frameworks like epistemic bubbles and misinformation diffusion, proposing a new language for understanding trust breaches and auditing inferential chains in densely interactive environments involving humans and large language models.