BOUND refines LLMs' package-validity boundary via targeted editing to cut package hallucination rates by 79.9% on edit prompts and 65.4% on unseen prompts in recommendation tasks while generalizing to code generation.
hub Canonical reference
A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Models
Canonical reference. 80% of citing Pith papers cite this work as background.
35
Pith papers citing it
114
external citations · Pith
Background
80% of classified citations
abstract
As Large Language Models (LLMs) continue to advance in their ability to write human-like text, a key challenge remains around their tendency to hallucinate generating content that appears factual but is ungrounded. This issue of hallucination is arguably the biggest hindrance to safely deploying these powerful LLMs into real-world production systems that impact people's lives. The journey toward widespread adoption of LLMs in practical settings heavily relies on addressing and mitigating hallucinations. Unlike traditional AI systems focused on limited tasks, LLMs have been exposed to vast amounts of online text data during training. While this allows them to display impressive language fluency, it also means they are capable of extrapolating information from the biases in training data, misinterpreting ambiguous prompts, or modifying the information to align superficially with the input. This becomes hugely alarming when we rely on language generation capabilities for sensitive applications, such as summarizing medical records, financial analysis reports, etc. This paper presents a comprehensive survey of over 32 techniques developed to mitigate hallucination in LLMs. Notable among these are Retrieval Augmented Generation (Lewis et al, 2021), Knowledge Retrieval (Varshney et al,2023), CoNLI (Lei et al, 2023), and CoVe (Dhuliawala et al, 2023). Furthermore, we introduce a detailed taxonomy categorizing these methods based on various parameters, such as dataset utilization, common tasks, feedback mechanisms, and retriever types. This classification helps distinguish the diverse approaches specifically designed to tackle hallucination issues in LLMs. Additionally, we analyze the challenges and limitations inherent in these techniques, providing a solid foundation for future research in addressing hallucinations and related phenomena within the realm of LLMs.
hub tools
citation-role summary
background 4
method 1
citation-polarity summary
representative citing papers
Activation patching reveals that citation decisions in Llama-3.1-8B RAG are implemented by a distributed attributional ensemble of heads and layers; targeted interventions fix most missed and spurious citations on PopQA.
An empirical taxonomy of 11 top-level categories and 27 subcategories of runtime faults in MCP servers, derived via open coding of GitHub threads and validated by a survey of 55 developers.
Introduces CHARM framework that detects cascading hallucinations in agentic RAG at 89.4% rate with 5.3% false positives and reduces error propagation by 82.1% on multi-hop QA benchmarks.
LLMs struggle to associate epistemic markers with stable internal confidence levels across distributions, even under model-centric interpretations, while maintaining somewhat consistent marker rankings.
VideoNet is a new large-scale benchmark and training dataset for domain-specific action recognition that exposes limitations in VLMs and enables smaller fine-tuned models to surpass larger open-weight ones.
HAVEN combines LLM agents for planning and gap analysis with protocol-specific templates and a custom DSL to generate correct UVM testbenches, achieving 100% compilation success, 90.6% code coverage, and 87.9% functional coverage on 19 open-source designs across three protocols.
This paper introduces a systems-level conceptual framing and a three-level taxonomy (intra-model, system-level, socio-technical) for uncertainty propagation in compound LLM applications, along with engineering insights and open challenges.
Context-Fidelity Boosting reduces faithfulness hallucinations by applying context-based logit boosts to source-supported tokens during LLM decoding.
GraphScout trains LLMs to autonomously synthesize structured training data from knowledge graphs via flexible exploration tools, enabling a 4B model to outperform larger LLMs by 16.7% on average with fewer inference tokens and strong cross-domain transfer.
SHIFT reformulates neuron editing as learnable gate modulation on under 0.01% parameters to let LLMs adaptively balance contextual and parametric knowledge during RAG generation.
A new framework quantifies faithful confidence expression in large reasoning models by comparing linguistic decisiveness to token probabilities, hidden states, and response consistency, revealing it as a persistent challenge.
REALISTA generates semantically coherent adversarial prompts via latent-space optimization over input-dependent editing directions, achieving stronger hallucination elicitation than prior realistic attacks on open-source and reasoning LLMs.
CiteTracer detects hallucinated citations with a 12-code field-level taxonomy and a cascading multi-agent retrieval pipeline, reporting 97.1% accuracy on a synthetic benchmark.
CAGE uses common-agency games and an EPEC algorithm to compute equilibrium policies that balance multiple conflicting objectives for test-time LLM alignment.
DisAAD trains a 1%-sized proxy model via adversarial distillation to quantify uncertainty in black-box LLMs by aligning with their output distributions.
Each tested LLM shows its own characteristic unreliability when engaging in repair during extended math-question dialogues.
PRISM benchmark disentangles LLM hallucinations into knowledge missing, knowledge errors, reasoning errors, and instruction-following errors across three generation stages, revealing trade-offs when testing 24 models.
LGCD creates pseudo-overlapping user data via LLM reasoning and uses conditional diffusion to generate target-domain user representations for inter-domain sequential recommendation without real overlapping users.
An importance-aware recall metric for LLM factuality evaluation reveals models are better at avoiding false claims than covering all relevant facts.
CiteAudit supplies a human-validated benchmark and multi-agent verification system that outperforms existing LLMs and commercial tools at detecting hallucinated scientific references.
UVM^2 is an LLM-driven system that generates and refines UVM testbenches for RTL verification, reporting up to substantial time savings and average code/function coverage of 87.44%/89.58% on designs up to 1.6K lines, outperforming prior methods.
CRAG improves RAG robustness via a retrieval quality evaluator that triggers web augmentation and a decompose-recompose filter to focus on relevant information, yielding better results on short- and long-form generation tasks.
NTS-CoT applies Element-CoT, Date Selection, and Causal-CoT modules to cut unfaithful content and omissions in LLM-based timeline summarization, outperforming baselines on three benchmarks.
citing papers explorer
-
Mitigating Package Hallucinations in Large Language Models via Model Editing
BOUND refines LLMs' package-validity boundary via targeted editing to cut package hallucination rates by 79.9% on edit prompts and 65.4% on unseen prompts in recommendation tasks while generalizing to code generation.
-
How Do LLMs Cite? A Mechanistic Interpretation of Attribution in Retrieval-Augmented Generation
Activation patching reveals that citation decisions in Llama-3.1-8B RAG are implemented by a distributed attributional ensemble of heads and layers; targeted interventions fix most missed and spurious citations on PopQA.
-
A Taxonomy of Runtime Faults in Model Context Protocol Servers
An empirical taxonomy of 11 top-level categories and 27 subcategories of runtime faults in MCP servers, derived via open coding of GitHub threads and validated by a survey of 55 developers.
-
Cascading Hallucination in Agentic RAG: The CHARM Framework for Detection and Mitigation
Introduces CHARM framework that detects cascading hallucinations in agentic RAG at 89.4% rate with 5.3% false positives and reduces error propagation by 82.1% on multi-hop QA benchmarks.
-
Can LLMs Use Linguistic Uncertainty Markers to Reliably Reflect Intrinsic Confidence?
LLMs struggle to associate epistemic markers with stable internal confidence levels across distributions, even under model-centric interpretations, while maintaining somewhat consistent marker rankings.
-
VideoNet: A Large-Scale Dataset for Domain-Specific Action Recognition
VideoNet is a new large-scale benchmark and training dataset for domain-specific action recognition that exposes limitations in VLMs and enables smaller fine-tuned models to surpass larger open-weight ones.
-
HAVEN: Hybrid Automated Verification ENgine for UVM Testbench Synthesis with LLMs
HAVEN combines LLM agents for planning and gap analysis with protocol-specific templates and a custom DSL to generate correct UVM testbenches, achieving 100% compilation success, 90.6% code coverage, and 87.9% functional coverage on 19 open-source designs across three protocols.
-
Uncertainty Propagation in LLM-Based Systems
This paper introduces a systems-level conceptual framing and a three-level taxonomy (intra-model, system-level, socio-technical) for uncertainty propagation in compound LLM applications, along with engineering insights and open challenges.
-
Context-Fidelity Boosting: Enhancing Faithful Generation through Watermark-Inspired Decoding
Context-Fidelity Boosting reduces faithfulness hallucinations by applying context-based logit boosts to source-supported tokens during LLM decoding.
-
GraphScout: Empowering Large Language Models with Intrinsic Exploration Ability for Agentic Graph Reasoning
GraphScout trains LLMs to autonomously synthesize structured training data from knowledge graphs via flexible exploration tools, enabling a 4B model to outperform larger LLMs by 16.7% on average with fewer inference tokens and strong cross-domain transfer.
-
SHIFT: Gate-Modulated Activation Steering for Knowledge Conflict Mitigation in Retrieval-Augmented Generation
SHIFT reformulates neuron editing as learnable gate modulation on under 0.01% parameters to let LLMs adaptively balance contextual and parametric knowledge during RAG generation.
-
Quantifying Faithful Confidence Expression in Large Reasoning Models
A new framework quantifies faithful confidence expression in large reasoning models by comparing linguistic decisiveness to token probabilities, hidden states, and response consistency, revealing it as a persistent challenge.
-
REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations
REALISTA generates semantically coherent adversarial prompts via latent-space optimization over input-dependent editing directions, achieving stronger hallucination elicitation than prior realistic attacks on open-source and reasoning LLMs.
-
Source or It Didn't Happen: A Multi-Agent Framework for Citation Hallucination Detection
CiteTracer detects hallucinated citations with a 12-code field-level taxonomy and a cascading multi-agent retrieval pipeline, reporting 97.1% accuracy on a synthetic benchmark.
-
Common-agency Games for Multi-Objective Test-Time Alignment
CAGE uses common-agency games and an EPEC algorithm to compute equilibrium policies that balance multiple conflicting objectives for test-time LLM alignment.
-
Estimating the Black-box LLM Uncertainty with Distribution-Aligned Adversarial Distillation
DisAAD trains a 1%-sized proxy model via adversarial distillation to quantify uncertainty in black-box LLMs by aligning with their output distributions.
-
Talking to a Know-It-All GPT or a Second-Guesser Claude? How Repair reveals unreliable Multi-Turn Behavior in LLMs
Each tested LLM shows its own characteristic unreliability when engaging in repair during extended math-question dialogues.
-
PRISM: Probing Reasoning, Instruction, and Source Memory in LLM Hallucinations
PRISM benchmark disentangles LLM hallucinations into knowledge missing, knowledge errors, reasoning errors, and instruction-following errors across three generation stages, revealing trade-offs when testing 24 models.
-
From Clues to Generation: Language-Guided Conditional Diffusion for Cross-Domain Recommendation
LGCD creates pseudo-overlapping user data via LLM reasoning and uses conditional diffusion to generate target-domain user representations for inter-domain sequential recommendation without real overlapping users.
-
Beyond Precision: Importance-Aware Recall for Factuality Evaluation in Long-Form LLM Generation
An importance-aware recall metric for LLM factuality evaluation reveals models are better at avoiding false claims than covering all relevant facts.
-
CiteAudit: You Cited It, But Did You Read It? A Benchmark for Verifying Scientific References in the LLM Era
CiteAudit supplies a human-validated benchmark and multi-agent verification system that outperforms existing LLMs and commercial tools at detecting hallucinated scientific references.
-
From Concept to Practice: an Automated LLM-aided UVM Machine for RTL Verification
UVM^2 is an LLM-driven system that generates and refines UVM testbenches for RTL verification, reporting up to substantial time savings and average code/function coverage of 87.44%/89.58% on designs up to 1.6K lines, outperforming prior methods.
-
Corrective Retrieval Augmented Generation
CRAG improves RAG robustness via a retrieval quality evaluator that triggers web augmentation and a decompose-recompose filter to focus on relevant information, yielding better results on short- and long-form generation tasks.
-
NTS-CoT: Mitigating Hallucinations in LLM-based News Timeline Summarization with Chain-of-Thought Reasoning
NTS-CoT applies Element-CoT, Date Selection, and Causal-CoT modules to cut unfaithful content and omissions in LLM-based timeline summarization, outperforming baselines on three benchmarks.
-
Hallucination Cascade: Analyzing Error Propagation in Multi-Agent LLM Systems
Experiments reveal that multi-agent LLM cascades reduce hallucination scores but also slightly decrease factual accuracy.
-
Restructure This: Using AI to Restructure Onboarding Documents to Reduce Cognitive Overload
VisDoc uses a GenAI pipeline grounded in CTML to restructure OSS onboarding docs, with small evaluations showing higher task success and lower cognitive load.
-
Teaching Astronomy with Large Language Models
Structured integration of LLMs in astronomy education, including a domain-specific tutor and documentation requirements, leads to improved AI literacy and reduced student reliance on AI over the semester.
-
Hallucination Detection and Evaluation of Large Language Model
HHEM delivers fast hallucination detection in LLMs via classification, cutting evaluation time from 8 hours to 10 minutes with up to 82.2% accuracy while adding segment retrieval for summarization.
-
An Insight into Security Code Review with LLMs: Capabilities, Obstacles, and Influential Factors
Empirical study finds LLMs outperform static analysis tools for security defect detection in code review, with DeepSeek-R1 best under commit-message plus chain-of-thought prompts.
-
K-Quantization and its Impact on Output Performance
Empirical evaluation of quantization effects on eight LLMs across bit widths, showing performance generally declines at lower precision but with model-size-dependent resilience and acceptable accuracy at 2 bits for many cases.
-
The Dynamic Gist-Based Memory Model (DGMM): A Memory-Centric Architecture for Artificial Intelligence
DGMM is proposed as an explicit graph-structured memory architecture for AI that enables persistent episodic memory, cue-based recall, and context-dependent interpretation without retraining.
-
LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methods
A survey that organizes LLMs-as-judges research into functionality, methodology, applications, meta-evaluation, and limitations.
-
A Survey on the Memory Mechanism of Large Language Model based Agents
A systematic review of memory designs, evaluation methods, applications, limitations, and future directions for LLM-based agents.
-
A Systematic Survey of Prompt Engineering in Large Language Models: Techniques and Applications
A systematic survey categorizes prompt engineering methods for LLMs and VLMs by application area, summarizing methodologies, applications, models, datasets, strengths, and limitations for each technique along with a taxonomy and summary table.
- When LLMs Invent Rust Crates: An Empirical Study of Hallucination Patterns and Mitigation