Pith. sign in

REVIEW 25 cited by

INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2402.03744 v2 pith:G4FO4KAW submitted 2024-02-06 cs.CL

Chao Chen , Kai Liu , Ze Chen , Yi Gu , Yue Wu , Mingyuan Tao , Zhihang Fu , Jieping Ye This is my paper · ORCID
classification cs.CL
keywords textbfllmsdetectionhallucinationsemanticdensehallucinationsinformation
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Knowledge hallucination have raised widespread concerns for the security and reliability of deployed LLMs. Previous efforts in detecting hallucinations have been employed at logit-level uncertainty estimation or language-level self-consistency evaluation, where the semantic information is inevitably lost during the token-decoding procedure. Thus, we propose to explore the dense semantic information retained within LLMs' \textbf{IN}ternal \textbf{S}tates for halluc\textbf{I}nation \textbf{DE}tection (\textbf{INSIDE}). In particular, a simple yet effective \textbf{EigenScore} metric is proposed to better evaluate responses' self-consistency, which exploits the eigenvalues of responses' covariance matrix to measure the semantic consistency/diversity in the dense embedding space. Furthermore, from the perspective of self-consistent hallucination detection, a test time feature clipping approach is explored to truncate extreme activations in the internal states, which reduces overconfident generations and potentially benefits the detection of overconfident hallucinations. Extensive experiments and ablation studies are performed on several popular LLMs and question-answering (QA) benchmarks, showing the effectiveness of our proposal.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 25 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. SpanUQ: Span-Level Uncertainty Quantification for Large Language Model Generation

    cs.CL 2026-07 conditional novelty 7.0 of 10

    A DETR-style probe distills multi-sample claim uncertainty into single-pass span detection and continuous Mixture-of-Beta scores, outperforming baselines on a new 293K-span benchmark.

  2. Clotho: Measuring Task-Specific Pre-Generation Test Adequacy for LLM Inputs

    cs.SE 2025-09 unverdicted novelty 7.0 of 10

    Clotho ranks LLM test inputs by failure likelihood using pre-generation hidden states and GMMs, achieving 0.716 ROC-AUC after labeling 5.4% of inputs on average across eight tasks and three models, with transfer to pr...

  3. TruthLens: Object Hallucination Detection via Self-Evaluating Truthfulness Scores in LVLMs

    cs.CV 2026-08 conditional novelty 6.0 of 10

    TruthLens fine-tunes LVLMs so the log-probability of a special token at each object mention becomes a truthfulness score, detecting object hallucinations with state-of-the-art AUROC.

  4. Confidently Wrong: Detecting Hallucinations in Financial Question Answering from LLM Internal States

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Among 8/8 self-consistent answers on FinQA, residual-stream probes detect wrong answers at 0.68–0.77 AUROC versus 0.55–0.63 for the best cheap output baselines across three 8–9B models.

  5. Hallucination Detection in Large Language Models Using Diversion Decoding

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Forcing an LLM away from its greedy answer yields resistance features that train a classifier detecting hallucinations more accurately and cheaply than semantic entropy.

  6. Reliability Scales Inversely: Hallucinations Snowball Faster in Bigger Language Models

    cs.LG 2026-06 conditional novelty 6.0 of 10

    Hallucination in LLMs is driven by an oracle-invisible “decoding risk” term that grows with scale and causally compounds errors within a response.

  7. Neural Message-Passing on Attention Graphs for Hallucination Detection

    cs.LG 2025-09 conditional novelty 6.0 of 10

    CHARM trains graph neural networks on token-attention graphs built from LLM computational traces and outperforms prior hallucination detectors on five benchmarks at token and response level.

  8. Unsupervised Hallucination Detection by Inspecting Reasoning Processes

    cs.CL 2025-09 conditional novelty 6.0 of 10

    IRIS detects LLM hallucinations by training a lightweight probe on hidden states elicited during the model's own step-by-step verification, using the model's verbalized confidence as soft pseudolabels.

  9. Rethinking LLM Parametric Knowledge as Post-retrieval Confidence for Dynamic Retrieval and Reranking

    cs.IR 2025-09 conditional novelty 6.0 of 10

    Shifts in an LLM's hidden-state confidence, before and after a retrieved context, are used as a preference signal to fine-tune a reranker and to trigger retrieval only when initial confidence is low.

  10. Decoding Memories: An Efficient Pipeline for Self-Consistency Hallucination Detection

    cs.CL 2025-08 conditional novelty 6.0 of 10

    A decoding pipeline reuses cached tokens and anneals sampling temperature to accelerate self-consistency hallucination detection by up to 3x without meaningful AUROC loss.

  11. ICR Probe: Tracking Hidden State Dynamics for Reliable Hallucination Detection in LLMs

    cs.CL 2025-07 conditional novelty 6.0 of 10

    A residual-stream update metric, the ICR Score, and a lightweight probe built on it detect LLM hallucinations with higher AUROC than several baselines across three open models.

  12. Shaking to Reveal: Perturbation-Based Detection of LLM Hallucinations

    cs.AI 2025-06 conditional novelty 6.0 of 10

    SSP adds a learned, sample-specific noise prompt to an LLM input and scores hallucination by the cosine shift in intermediate representations, outperforming output-confidence baselines on QA benchmarks.

  13. Beyond Semantic Entropy: Boosting LLM Uncertainty Quantification with Pairwise Semantic Similarity

    cs.LG 2025-05 conditional novelty 6.0 of 10

    SNNE estimates LLM uncertainty from pairwise semantic similarities of sampled answers using a log-sum-exp aggregation, and it generalizes semantic entropy as a special case.

  14. Are the Hidden States Hiding Something? Testing the Limits of Factuality-Encoding Capabilities in LLMs

    cs.CL 2025-05 conditional novelty 6.0 of 10

    Hidden-state factuality probes trained on synthetic statements do not generalize to LLM-generated factual statements, despite reproducing prior results on original datasets.

  15. DUD: Decoupled Update Dynamics for Reliable Uncertainty Quantification in Large Language Models

    cs.CL 2026-08 conditional novelty 5.0 of 10

    DUD decouples Attention and FFN contributions via noise-induced causal patching and trains a probe on the dual-stream restoration scores, reporting state-of-the-art hallucination detection.

  16. Scene Generation at Absolute Scale: Utilizing Semantic and Geometric Guidance From Text for Accurate and Interpretable 3D Indoor Scene Generation

    cs.CV 2026-03 unverdicted novelty 5.0 of 10

    A text-to-3D pipeline that predicts a global semantic-geometric layout, conditions panoramic and video diffusion on it, and fuses views with 3D Gaussian Splatting into an absolute-scale navigable indoor scene.

  17. Entropy Sentinel: Probing Entropy Traces for LLM Monitoring

    cs.CL 2026-01 conditional novelty 5.0 of 10

    Top-k decoding-entropy profiles can estimate and rank held-out domain accuracy for most tested LLMs, with difficulty-diverse training data the main success factor.

  18. ISACL: Internal State Analyzer for Copyrighted Training Data Leakage

    cs.CL 2025-08 conditional novelty 5.0 of 10

    An MLP trained on LLM internal states predicts Rouge-L-defined literal copying leakage with high accuracy, but not paraphrase-level leakage.

  19. Cleanse: Uncertainty Estimation Approach Using Clustering-based Semantic Consistency in LLMs

    cs.CL 2025-07 conditional novelty 5.0 of 10

    Cleanse detects hallucinated LLM answers by computing the share of hidden-embedding cosine similarity that falls inside semantic clusters, and it beats several baselines in AUROC across four models and two QA benchmarks.

  20. Do BERT-Like Bidirectional Models Still Perform Better on Text Classification in the Era of LLMs?

    cs.CL 2025-05 conditional novelty 5.0 of 10

    Fine-tuned BERT-like models outperform zero-shot and internal-state LLM methods on four of six challenging text classification datasets.

  21. Shadows in the Attention: Contextual Perturbation and Representation Drift in the Dynamics of Hallucination in LLMs

    cs.CL 2025-05 reject novelty 5.0 of 10

    Context injection shifts LLM hidden states and attention, and the paper claims this drift tracks and eventually "locks in" hallucination behavior, but the reported data contain internal inconsistencies.

  22. Beyond ROUGE: N-Gram Subspace Features for LLM Hallucination Detection

    cs.CL 2025-09 reject novelty 4.0 of 10

    Singular values of label-grouped n-gram frequency tensors are used as MLP features for hallucination detection, with reported gains on HaluEval that rely on label-aware grouping.

  23. How Knowledge Popularity Influences and Enhances LLM Knowledge Boundary Perception

    cs.CL 2025-05 conditional novelty 4.0 of 10

    Entity popularity and entity co-occurrence in Wikipedia correlate with LLM QA accuracy, confidence, and calibration, and combining them with confidence improves answer-correctness prediction by 5.24% on average.

  24. Loki's Dance of Illusions: A Comprehensive Survey of Hallucination in Large Language Models

    cs.CL 2025-06 reject novelty 3.0 of 10

    A survey of LLM hallucination research that formalizes hallucination types and argues, via incompleteness and undecidability arguments, that hallucinations cannot be fully eliminated.

  25. From Hallucinations to Jailbreaks: Rethinking the Vulnerability of Large Foundation Models

    cs.CV 2025-05 reject novelty 3.0 of 10

    The paper argues hallucinations and jailbreaks share the same optimization dynamics and shows that defenses for one also reduce the other, but the theoretical support is largely circular.

Pith tools