A new benchmark with cognitive traps shows frontier deep research agents achieve only 13-16% acceptance on expert consulting tasks under combined verifier and rubric criteria.
hub Canonical reference
A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions
Canonical reference. 85% of citing Pith papers cite this work as background.
77
Pith papers citing it
215
external citations · Pith
Background
85% of classified citations
abstract
The emergence of large language models (LLMs) has marked a significant breakthrough in natural language processing (NLP), fueling a paradigm shift in information acquisition. Nevertheless, LLMs are prone to hallucination, generating plausible yet nonfactual content. This phenomenon raises significant concerns over the reliability of LLMs in real-world information retrieval (IR) systems and has attracted intensive research to detect and mitigate such hallucinations. Given the open-ended general-purpose attributes inherent to LLMs, LLM hallucinations present distinct challenges that diverge from prior task-specific models. This divergence highlights the urgency for a nuanced understanding and comprehensive overview of recent advances in LLM hallucinations. In this survey, we begin with an innovative taxonomy of hallucination in the era of LLM and then delve into the factors contributing to hallucinations. Subsequently, we present a thorough overview of hallucination detection methods and benchmarks. Our discussion then transfers to representative methodologies for mitigating LLM hallucinations. Additionally, we delve into the current limitations faced by retrieval-augmented LLMs in combating hallucinations, offering insights for developing more robust IR systems. Finally, we highlight the promising research directions on LLM hallucinations, including hallucination in large vision-language models and understanding of knowledge boundaries in LLM hallucinations.
hub tools
citation-role summary
background 13
citation-polarity summary
roles
background 13representative citing papers
Models delayed verification in multi-agent LLMs as graph consensus, derives stability thresholds (inverse golden ratio for delay two) via grounded Laplacian, and gives a supermodular greedy rule for corrector placement; experiments on five models confirm dose-delay oscillations.
MM-Snowball benchmark diagnoses hallucination snowballing in multi-turn MLLM dialogues; CAVR mitigates it via dual visual rectification at representation and logit levels.
Hallucination is detected as a transport-cost excursion in hidden-state trajectories, localized via contrastive PCA in a teacher model and distilled to a BiLSTM student.
Introduces TBPO, which derives a Bregman-divergence density-ratio matching objective for token-level preference optimization that generalizes DPO while preserving the induced optimal policy.
GSAR is a grounding-evaluation framework for multi-agent LLMs that uses a four-way claim typology, evidence-weighted asymmetric scoring, and tiered recovery decisions to detect and mitigate hallucinations.
Activation probes detect hallucinations pre-generation in large LLMs but cannot correct them via steering, with output confidence outperforming on accuracy.
LatentRefusal predicts answerability of text-to-SQL queries from LLM hidden states using a Tri-Residual Gated Encoder, reaching 88.5% average F1 across four benchmarks with about 2ms overhead.
A study of seven LLMs finds that realistic prompt variations such as one-character misspellings trigger library hallucinations in up to 26% of cases, fabricated names in up to 99%, and time-based prompts in up to 85%, and introduces LibHalluBench for evaluation.
Introduces Trust-RAG Compass framework and TRC Bench benchmark to assess RAG trustworthiness across factuality, robustness, fairness, transparency, accountability, and privacy, with evaluations showing performance gaps between LLMs.
Hallucinations are inevitable in LLMs because they cannot learn all computable functions according to learning theory.
HQRE entropy regularization makes multi-agent LLM coordination well-posed, yielding unique equilibria, linear mirror convergence, bounded Bayesian regret, and DICE gains of 4.3–8.5 pp on reasoning/planning tasks.
A learned weighted sum of intermediate-layer activations compresses an instruction prompt into a single patch vector that, injected at an early layer, recovers task accuracy within ~2% of the full prompt.
Global calibration metrics like ECE are confounded by accuracy; the proposed ACE framework with three accuracy-controlled views shows many prior calibration advantages weaken or reverse.
Mandatory per-line citations in SDD frameworks reduce LLM output determinism but enable reliable automated hallucination detection (TDR 86-88%, FPR 0%), a trade-off replicated across Claude and GLM models.
A small parametric transition model plus a Jaccard consistency gate grounds LLM agent state deltas, cutting hallucinated-state rate ~80% and raising success from 0.668 to 0.838 on graph planning benchmarks.
A memory that keeps a wrong conclusion while dropping the source makes an LLM confidently repeat the error; keeping the source instead lets a later correction recover the truth at the same budget.
Autopilot enforces verifiable termination via a gated FSM scheduler and hard floor, proving that termination implies goal achievement under gate soundness, floor enforcement, and plan coverage, while cutting fabrication rates to 0.95% vs. 8-25% in baselines on 3150 paired cells including SWE-bench L
Factual knowledge crystallizes abruptly in final layers of LLMs (26.8%-93.4% of correct answers absent from top-10 until end), explaining why CAA outperforms DoLa on some models but not others.
EGC reveals that graph consistency measures align with hallucinations in Llama-2 but reverse direction in GPT-4, GPT-3.5 and Mistral-7B on the RAGTruth QA split, indicating model-family-specific hallucination patterns.
Grounded Decoding fuses full-RAG and retrieval-only next-token distributions via normalized geometric mean from a KL-barycenter to improve factual consistency and citation quality in RAG.
Introduces functional equivalence methods and functional entropy to predict functional correctness of LLM-generated code via uncertainty quantification, outperforming NLI-based baselines in most tested settings.
Benchmark construction artifacts in hallucination detection corpora allow naive text-similarity baselines to achieve near-perfect scores, and controlled evaluations show most methods perform near chance except SAPLMA and the new DRIFT probe.
Introduces a trust-boundary architecture in Lean 4 with three certificate families and two operators that deliver sorry-free, axiom-audited assurances for LLM pipeline components.
citing papers explorer
-
Evaluating Deep Research Agents on Expert Consulting Work: A Benchmark with Verifiers, Rubrics, and Cognitive Traps
A new benchmark with cognitive traps shows frontier deep research agents achieve only 13-16% acceptance on expert consulting tasks under combined verifier and rubric criteria.
-
Delayed Verification Destabilizes Multi-Agent LLM Belief: Instability Thresholds and Optimal Corrector Placement
Models delayed verification in multi-agent LLMs as graph consensus, derives stability thresholds (inverse golden ratio for delay two) via grounded Laplacian, and gives a supermodular greedy rule for corrector placement; experiments on five models confirm dose-delay oscillations.
-
MM-Snowball: Evaluating and Mitigating Hallucination Snowballing in Multimodal Multi-Turn Dialogue
MM-Snowball benchmark diagnoses hallucination snowballing in multi-turn MLLM dialogues; CAVR mitigates it via dual visual rectification at representation and logit levels.
-
Where Does Reasoning Break? Step-Level Hallucination Detection via Hidden-State Transport Geometry
Hallucination is detected as a transport-cost excursion in hidden-state trajectories, localized via contrastive PCA in a teacher model and distilled to a BiLSTM student.
-
TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching
Introduces TBPO, which derives a Bregman-divergence density-ratio matching objective for token-level preference optimization that generalizes DPO while preserving the induced optimal policy.
-
GSAR: Typed Grounding for Hallucination Detection and Recovery in Multi-Agent LLMs
GSAR is a grounding-evaluation framework for multi-agent LLMs that uses a four-way claim typology, evidence-weighted asymmetric scoring, and tiered recovery decisions to detect and mitigate hallucinations.
-
Detection Without Correction: A Robust Asymmetry in Activation-Based Hallucination Probing
Activation probes detect hallucinations pre-generation in large LLMs but cannot correct them via steering, with output confidence outperforming on accuracy.
-
LatentRefusal: Latent-Signal Refusal for Unanswerable Text-to-SQL Queries
LatentRefusal predicts answerability of text-to-SQL queries from LLM hidden states using a Tri-Residual Gated Encoder, reaching 88.5% average F1 across four benchmarks with about 2ms overhead.
-
Library Hallucinations in LLM-Generated Code: A Risk Analysis Grounded in Developer Queries
A study of seven LLMs finds that realistic prompt variations such as one-character misspellings trigger library hallucinations in up to 26% of cases, fabricated names in up to 99%, and time-based prompts in up to 85%, and introduces LibHalluBench for evaluation.
-
Trustworthiness in Retrieval-Augmented Generation Systems: A Survey
Introduces Trust-RAG Compass framework and TRC Bench benchmark to assess RAG trustworthiness across factuality, robustness, fairness, transparency, accountability, and privacy, with evaluations showing performance gaps between LLMs.
-
Hallucination is Inevitable: An Innate Limitation of Large Language Models
Hallucinations are inevitable in LLMs because they cannot learn all computable functions according to learning theory.
-
DICE: Entropy-Regularized Equilibrium Selection for Stable Multi-Agent LLM Coordination
HQRE entropy regularization makes multi-agent LLM coordination well-posed, yielding unique equilibria, linear mirror convergence, bounded Bayesian regret, and DICE gains of 4.3–8.5 pp on reasoning/planning tasks.
-
Prompt Compression via Activation Aggregation
A learned weighted sum of intermediate-layer activations compresses an instruction prompt into a single patch vector that, injected at an early layer, recovers task accuracy within ~2% of the full prompt.
-
When Calibration Rankings Reverse: Accuracy-Controlled Evaluation for Fair Comparison of LLMs
Global calibration metrics like ECE are confounded by accuracy; the proposed ACE framework with three accuracy-controlled views shows many prior calibration advantages weaken or reverse.
-
Citation Discipline in Spec-Driven Development: A Cross-Model Empirical Study of Output Determinism and Automated Hallucination Detection in LLM-Generated Code
Mandatory per-line citations in SDD frameworks reduce LLM output determinism but enable reliable automated hallucination detection (TDR 86-88%, FPR 0%), a trade-off replicated across Claude and GLM models.
-
Agent vs. Parametric World Models: Hybrid Planning for Reliable Language Agents
A small parametric transition model plus a Jaccard consistency gate grounds LLM agent state deltas, cutting hallucinated-state rate ~80% and raising success from 0.668 to 0.838 on graph planning benchmarks.
-
Reclaim Evaluation: A Lossy Memory Is Worse Than an Empty One
A memory that keeps a wrong conclusion while dropping the source makes an LLM confidently repeat the error; keeping the source instead lets a later correction recover the truth at the same budget.
-
Goal-Autopilot: A Verifiable Anti-Fabrication Firewall for Unattended Long-Horizon Agents
Autopilot enforces verifiable termination via a gated FSM scheduler and hard floor, proving that termination implies goal achievement under gate soundness, floor enforcement, and plan coverage, while cutting fabrication rates to 0.95% vs. 8-25% in baselines on 3150 paired cells including SWE-bench L
-
MechLens: Late Crystallization of Factual Knowledge Explains Intervention Effectiveness in Language Models
Factual knowledge crystallizes abruptly in final layers of LLMs (26.8%-93.4% of correct answers absent from top-10 until end), explaining why CAA outperforms DoLa on some models but not others.
-
Evidence Graph Consistency in Retrieval-Augmented Generation: A Model-Dependent Analysis of Hallucination Detection
EGC reveals that graph consistency measures align with hallucinations in Llama-2 but reverse direction in GPT-4, GPT-3.5 and Mistral-7B on the RAGTruth QA split, indicating model-family-specific hallucination patterns.
-
Grounded Decoding: Retrieval-Anchored Probability Fusion for Faithful RAG
Grounded Decoding fuses full-RAG and retrieval-only next-token distributions via normalized geometric mean from a KL-barycenter to improve factual consistency and citation quality in RAG.
-
Functional Entropy: Predicting Functional Correctness in LLM-Generated Code with Uncertainty Quantification
Introduces functional equivalence methods and functional entropy to predict functional correctness of LLM-generated code via uncertainty quantification, outperforming NLI-based baselines in most tested settings.
-
PARALLAX: Separating Genuine Hallucination Detection from Benchmark Construction Artifacts
Benchmark construction artifacts in hallucination detection corpora allow naive text-similarity baselines to achieve near-perfect scores, and controlled evaluations show most methods perform near chance except SAPLMA and the new DRIFT probe.
-
Proof-Carrying Certificates for LLM Pipelines: A Trust-Boundary Architecture
Introduces a trust-boundary architecture in Lean 4 with three certificate families and two operators that deliver sorry-free, axiom-audited assurances for LLM pipeline components.
-
Template-as-Ontology: Configurable Synthetic Data Infrastructure for Cross-Domain Manufacturing AI Validation
A single configuration file generates causally coherent synthetic MES data across domains and guarantees zero tool-parameter hallucination when AI tools are ontology-constrained.
-
Low-Cost Black-Box Detection of LLM Hallucinations via Dynamical System Prediction
A single-pass black-box method models LLM outputs as dynamical systems via Koopman operators to detect hallucinations with claimed state-of-the-art accuracy and lower cost.
-
Quantization Undoes Alignment: Bias Emergence in Compressed LLMs Across Models and Precision Levels
3-bit quantization induces new stereotypical biases in 6-21% of previously unbiased BBQ items across three LLMs, undetected by perplexity increases under 3%, with models declining in 'unknown' responses by 17.4%.
-
The Surprising Universality of LLM Outputs: A Real-Time Verification Primitive
LLM token rank-frequency distributions converge to a shared Mandelbrot distribution across models and domains, enabling a microsecond-scale statistical primitive for provenance verification and black-box anomaly triage.
-
The Provenance Gap in Clinical AI: Evidence-Traceable Temporal Knowledge Graphs for Rare Disease Reasoning
HEG-TKG grounds LLM clinical reasoning in hierarchical evidence-based temporal knowledge graphs from 4,512 PubMed records, delivering 100% citation verifiability and error detectability where standard RAG and unprompted LLMs produce none.
-
When Agents Go Quiet: Output Generation Capacity and Format-Cost Separation for LLM Document Synthesis
LLM agents avoid output stalling and reduce generation tokens by 48-72% via deferred template rendering guided by Output Generation Capacity and a Format-Cost Separation Theorem.
-
EvoRAG: Making Knowledge Graph-based RAG Automatically Evolve through Feedback-driven Backpropagation
EvoRAG adds a feedback-driven backpropagation step that attributes response quality to individual knowledge-graph triplets and updates the graph to raise reasoning accuracy by 7.34 percent over prior KG-RAG methods.
-
Hallucination as Trajectory Commitment: Causal Evidence for Asymmetric Attractor Dynamics in Transformer Generation
Hallucination is an early trajectory commitment in transformers governed by asymmetric attractor dynamics, with prompt encoding selecting the basin and correction needing multi-step intervention.
-
FocalLens: Visualizing Narratives through Focalization
FocalLens is a new visualization system that captures focalization to display character perceptions, direct/indirect involvement, and narration in narratives, evaluated qualitatively with writers and scholars.
-
Phase-Associative Memory: Sequence Modeling in Complex Hilbert Space
PAM, a complex-valued associative memory model, exhibits steeper power-law scaling in loss and perplexity than a matched real-valued baseline when trained on WikiText-103 from 5M to 100M parameters.
-
Do Hallucination Neurons Generalize? Evidence from Cross-Domain Transfer in LLMs
Hallucination neurons in LLMs are domain-specific, with cross-domain classifiers dropping from AUROC 0.783 within-domain to 0.563 across domains.
-
Locate-then-Sparsify: Attribution Guided Sparse Strategy for Visual Hallucination Mitigation
LTS-FS locates hallucination-relevant layers in LVLMs via causal attribution on a constructed dataset and applies sparse layerwise feature steering to mitigate hallucinations while preserving general task performance.
-
Hallucination as output-boundary misclassification: a composite abstention architecture for language models
A composite abstention architecture combining instruction prompting and a support-deficit gate from three black-box signals reduces hallucinations more effectively than either component alone across tested models and regimes.
-
GhostCite: A Large-Scale Analysis of Citation Validity in the Age of Large Language Models
LLMs hallucinate citations at rates from 14.23% to 94.93%, with 1.07% of papers containing invalid citations and an 80.9% increase in 2025.
-
A Geometric Taxonomy of Hallucinations in LLMs
Embedding geometry on the unit hypersphere distinguishes detectable query-proximate unfaithfulness and confabulations from undetectable factual errors sharing vocabulary with correct answers.
-
A Retrieval-Augmented Generation Approach to Extracting Algorithmic Logic from Neural Networks
NN-RAG extracts 1,289 candidate neural modules from 19 PyTorch repositories, validates 941 of them, and supplies roughly 72% of the novel structures in the LEMUR dataset while enabling cross-repository migration.
-
RA-RRG: Multimodal Retrieval-Augmented Radiology Report Generation with Key Phrase Extraction
RA-RRG extracts key phrases with LLMs, retrieves them via multimodal similarity, and conditions report generation on them to achieve SOTA CheXbert scores and competitive RadGraph F1 on MIMIC-CXR and IU X-ray while supporting multi-view inputs.
-
In-depth Analysis of Graph-based RAG in a Unified Framework
A unified framework and large-scale comparison of graph-based RAG methods on QA tasks yields new high-performing variants obtained by recombining existing components.
-
Hallucinations are inevitable but can be made statistically negligible
Hallucinations are inevitable on an infinite set of inputs but can be made statistically negligible with sufficient training data quality and quantity.
-
A Red Teaming Framework for Large Language Models: A Case Study on Faithfulness Evaluation
A multi-role red-teaming framework with attacker, target, and jury LLMs measures faithfulness in English and Arabic, finding false-premise prompts and length limits change unfaithfulness rates.
-
Improving Collaborative Storytelling with a Multi-Agent Framework Based on Large Language Models
An iterative writer-editor multi-agent LLM process improves perceived story quality in simulations of child collaborative storytelling.
-
When Is Next-Token Prediction Useful? Marginalization, Ergodicity, Mixture Identifiability, Local Sufficiency, RAG, Tools, and Programming
Next-token prediction estimates a marginal text law that is useful only under ergodicity assumptions and when observed prefixes carry low residual mutual information about omitted latent circumstances.
-
Towards FairRAG: Preventing Representational Harm in Retrieval-Augmented Generation by Enforcing Fair Exposure at Retrieval Time
Representative Stochastic ranking achieves near-parity average exposure on the TREC 2022 Fair Ranking Dataset, with generation demographic parity closely tracking retrieval exposure.
-
The Semantic Training Gap: Ontology-Grounded Tool Architectures for Industrial AI Agent Systems
Ontology-grounded tool architectures eliminate hallucination of domain identifiers in industrial AI agents by enforcing semantic constraints through a typed relational configuration and three-operation interface.
-
EmoS: A High-Fidelity Multimodal Benchmark for Fine-grained Streaming Emotional Understanding
EmoS is a new high-fidelity benchmark for fine-grained streaming emotional understanding that produces measurable gains when used to fine-tune multimodal large language models.
-
CAST: Mitigating Object Hallucination in Large Vision-Language Models via Caption-Guided Visual Attention Steering
A training-free method that steers specific attention heads toward caption-query states reduces object hallucination in LVLMs by ~6% on average across five benchmarks.