Testing the Limits of Truth Directions in LLMs

• URL: http://arxiv.org/abs/2604.03754v1
• Date: Sat, 04 Apr 2026 15:01:09 GMT
• Title: Testing the Limits of Truth Directions in LLMs
• Abstract summary: We show that truth directions are highly layer-dependent, and that a full understanding of universality requires probing at many layers in the model.<n>We also show that truth directions depend heavily on task type, emerging in earlier layers for factual and later layers for reasoning tasks.<n>Our findings indicate that universality claims for truth directions are more limited than previously known, with significant differences observable for various model layers, task difficulties, task types, and prompt templates.
• Score: 8.253651182480981
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) have been shown to encode truth of statements in their activation space along a linear truth direction. Previous studies have argued that these directions are universal in certain aspects, while more recent work has questioned this conclusion drawing on limited generalization across some settings. In this work, we identify a number of limits of truth-direction universality that have not been previously understood. We first show that truth directions are highly layer-dependent, and that a full understanding of universality requires probing at many layers in the model. We then show that truth directions depend heavily on task type, emerging in earlier layers for factual and later layers for reasoning tasks; they also vary in performance across levels of task complexity. Finally, we show that model instructions dramatically affect truth directions; simple correctness evaluation instructions significantly affect the generalization ability of truth probes. Our findings indicate that universality claims for truth directions are more limited than previously known, with significant differences observable for various model layers, task difficulties, task types, and prompt templates.

Related papers

• Emergence of Linear Truth Encodings in Language Models [64.86571541830598]
  Large language models exhibit linear subspaces that separate true from false statements, yet the mechanism behind their emergence is unclear.<n>We introduce a transparent, one-layer transformer toy model that reproduces such truth subspaces end-to-end.<n>We study one simple setting in which truth encoding can emerge, encouraging the model to learn this distinction in order to lower the LM loss on future tokens.
  arXiv  Detail & Related papers  (2025-10-17T16:30:07Z)
• The Geometries of Truth Are Orthogonal Across Tasks [24.684427329854202]
  We show that "geometries of truth" are intrinsically task-dependent and fail to transfer across tasks.<n>More sophisticated approaches fail to overcome this limitation, likely because activation vectors commonly used to classify answers form clearly separated clusters when examined across tasks.
  arXiv  Detail & Related papers  (2025-06-10T08:40:31Z)
• Probing the Geometry of Truth: Consistency and Generalization of Truth Directions in LLMs Across Logical Transformations and Question Answering Tasks [31.379237532476875]
  We investigate whether large language models (LLMs) encode truthfulness as a distinct linear feature, termed the "truth direction"<n>Our findings reveal that not all LLMs exhibit consistent truth directions, with stronger representations observed in more capable models.<n>We show that truthfulness probes trained on declarative atomic statements can generalize effectively to logical transformations, question-answering tasks, in-context learning, and external knowledge sources.
  arXiv  Detail & Related papers  (2025-06-01T03:55:53Z)
• Exploring the generalization of LLM truth directions on conversational formats [0.0]
  We show that linear probes trained on a single hidden state of the model already generalize across a range of topics.<n>We find good generalization between short conversations that end on a lie, but poor generalization to longer formats where the lie appears earlier in the input prompt.
  arXiv  Detail & Related papers  (2025-05-14T21:21:08Z)
• Exploring the Generalizability of Factual Hallucination Mitigation via Enhancing Precise Knowledge Utilization [49.95746521480879]
  We introduce PKUE (Precise Knowledge Utilization Enhancement), which fine-tunes the model on self-generated responses to precise and simple factual questions.<n>Extensive experiments demonstrate that PKUE significantly improves LLM overall performance.
  arXiv  Detail & Related papers  (2025-02-26T13:34:52Z)
• Enhanced Language Model Truthfulness with Learnable Intervention and Uncertainty Expression [19.69104070561701]
  Large language models (LLMs) can generate long-form and coherent text, yet they often hallucinate facts. We propose LITO, a Learnable Intervention method for Truthfulness Optimization. Experiments on multiple LLMs and question-answering datasets demonstrate that LITO improves truthfulness while preserving task accuracy.
  arXiv  Detail & Related papers  (2024-05-01T03:50:09Z)
• The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets [6.732432949368421]
  Large Language Models (LLMs) have impressive capabilities, but are prone to outputting falsehoods. Recent work has developed techniques for inferring whether a LLM is telling the truth by training probes on the LLM's internal activations. We present evidence that at sufficient scale, LLMs linearly represent the truth or falsehood of factual statements.
  arXiv  Detail & Related papers  (2023-10-10T17:54:39Z)
• Do Large Language Models Know about Facts? [60.501902866946]
  Large language models (LLMs) have recently driven striking performance improvements across a range of natural language processing tasks. We aim to evaluate the extent and scope of factual knowledge within LLMs by designing the benchmark Pinocchio. Pinocchio contains 20K diverse factual questions that span different sources, timelines, domains, regions, and languages.
  arXiv  Detail & Related papers  (2023-10-08T14:26:55Z)
• DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models [79.01926242857613]
  Large language models (LLMs) are prone to hallucinations, generating content that deviates from facts seen during pretraining. We propose a simple decoding strategy for reducing hallucinations with pretrained LLMs. We find that this Decoding by Contrasting Layers (DoLa) approach is able to better surface factual knowledge and reduce the generation of incorrect facts.
  arXiv  Detail & Related papers  (2023-09-07T17:45:31Z)
• Leveraging sparse and shared feature activations for disentangled representation learning [112.22699167017471]
  We propose to leverage knowledge extracted from a diversified set of supervised tasks to learn a common disentangled representation. We validate our approach on six real world distribution shift benchmarks, and different data modalities.
  arXiv  Detail & Related papers  (2023-04-17T01:33:24Z)
• FaVIQ: FAct Verification from Information-seeking Questions [77.7067957445298]
  We construct a large-scale fact verification dataset called FaVIQ using information-seeking questions posed by real users. Our claims are verified to be natural, contain little lexical bias, and require a complete understanding of the evidence for verification.
  arXiv  Detail & Related papers  (2021-07-05T17:31:44Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.