mlsanityData to 14 Oct 2026

A Single Direction of Truth: An Observer Model's Linear Residual Probe Exposes and Steers Contextual Hallucinations

arXiv 2507.23221 ↗First posted 31 Jul 2025cs.LGcited by 9

Not on the map. Why?

Abstract (arXiv)

Contextual hallucinations -- statements unsupported by given context -- remain a significant challenge in AI. We demonstrate a practical interpretability insight: a generator-agnostic observer model detects hallucinations via a single forward pass and a linear probe on its residual stream. This probe isolates a single, transferable linear direction separating hallucinated from faithful text, outperforming baselines by 5-27 points and showing robust mid-layer performance across Gemma-2 models (2B to 27B). Gradient-times-activation localises this signal to sparse, late-layer MLP activity. Critically, manipulating this direction causally steers generator hallucination rates, proving its actionability. Our results offer novel evidence of internal, low-dimensional hallucination tracking linked to specific MLP sub-circuits, exploitable for detection and mitigation. We release the 2000-example ContraTales benchmark for realistic assessment of such solutions.

Key facts. Cited by 9 papers in the map (data through 14 Oct 2026). Builds on 28 of its 36 references that are in the map. First posted on arXiv 31 Jul 2025.

Builds on

28 of 36 references.

References · 28

  1. 2020The Pile: An 800GB Dataset of Diverse Text for Language Modelingcited by 2,411
  2. 2022Survey of Hallucination in Natural Language Generationcited by 2,075
  3. 2023A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questionscited by 1,601
  4. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  5. 2016Understanding intermediate layers using linear classifier probescited by 1,353
  6. 2018Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarizationcited by 1,317
  7. 2017Get To The Point: Summarization with Pointer-Generator Networkscited by 1,292
  8. 2022Toy Models of Superpositioncited by 971
  9. 2020On Faithfulness and Factuality in Abstractive Summarizationcited by 900
  10. 2023SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Modelscited by 810
  11. 2022Discovering Latent Knowledge in Language Models Without Supervisioncited by 767
  12. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 716
16 more
  1. 2023The Internal State of an LLM Knows When It's Lyingcited by 659
  2. 2024Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingcited by 529
  3. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 435
  4. 2023Emergent Linear Representations in World Models of Self-Supervised Sequence Modelscited by 319
  5. 2024Semantic Entropy Probes: Robust and Cheap Hallucination Detection in LLMscited by 259
  6. 2013k-Sparse Autoencoderscited by 215
  7. 2024Lookback Lens: Detecting and Mitigating Contextual Hallucinations in Large Language Models Using Only Attention Mapscited by 142
  8. 2021Entity-level Factual Consistency of Abstractive Text Summarizationcited by 118
  9. 2024A Practical Review of Mechanistic Interpretability for Transformer-Based Language Modelscited by 118
  10. 2024Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Modelscited by 115
  11. 2023Attention Satisfies: A Constraint-Satisfaction Lens on Factual Errors of Language Modelscited by 67
  12. 2023The Curious Case of Hallucinatory (Un)answerability: Finding Truths in the Hidden States of Over-Confident Large Language Modelscited by 55
  13. 2024Mechanistic Understanding and Mitigation of Language Model Non-Factual Hallucinationscited by 38
  14. 2024Controllable Context Sensitivity and the Knob Behind Itcited by 23
  15. 2024Constructing Benchmarks and Interventions for Combating Hallucinations in LLMscited by 19
  16. 2024Cost-Effective Hallucination Detection for LLMscited by 18

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 4; 2026: 5.

9 papers.

  1. 2025Beyond Transcription: Mechanistic Interpretability in ASRcited by 13
  2. 2026Steered LLM Activations are Non-Surjectivecited by 8
  3. 2025When Bias Pretends to Be Truth: How Spurious Correlations Undermine Hallucination Detection in LLMscited by 3
  4. 2026PARALLAX: Separating Genuine Hallucination Detection from Benchmark Construction Artifactscited by 3
  5. 2025Painless Activation Steering: An Automated, Lightweight Approach for Post-Training Large Language Modelscited by 2
  6. 2025The Unintended Trade-off of AI Alignment:Balancing Hallucination Mitigation and Safety in LLMscited by 2
  7. 2026Temporal Multi-Signal Fusion for Token-Level Hallucination Detectioncited by 1
  8. 2026Prediction of Prediction (PoP): Inter-Layer Activation Fusion for Single-Pass Hallucination Detection in Large Language Modelscited by 0
  9. 2026Extracting Persona Subspaces Through Iterative Nullspace Projection For Modulationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Detection Without Correction: A Robust Asymmetry in Activation-Based Hallucination Probing
  2. 2023Do Androids Know They're Only Dreaming of Electric Sheep?
  3. 2026External Observers May See More Clearly: Cross-Model Span-Level Hallucination Detection in Large Language Models via Hidden State Probing
  4. 2026HALP: Detecting Hallucinations in Vision-Language Models without Generating a Single Token
  5. 2026VIB-Probe: Detecting and Mitigating Hallucinations in Vision-Language Models via Variational Information Bottleneck
  6. 2025Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling
  7. 2026HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders
  8. 2025InterpDetect: Interpretable Signals for Detecting Hallucinations in Retrieval-Augmented Generation