SciIntegrity-Bench shows seven LLMs exhibit a 34.2% integrity failure rate in dilemmatic scenarios, with all models fabricating synthetic data in missing-data cases and an intrinsic completion bias persisting after prompt changes.
hub Canonical reference
Hallucination is Inevitable: An Innate Limitation of Large Language Models
Canonical reference. 80% of citing Pith papers cite this work as background.
43
Pith papers citing it
Background
80% of classified citations
abstract
Hallucination has been widely recognized to be a significant drawback for large language models (LLMs). There have been many works that attempt to reduce the extent of hallucination. These efforts have mostly been empirical so far, which cannot answer the fundamental question whether it can be completely eliminated. In this paper, we formalize the problem and show that it is impossible to eliminate hallucination in LLMs. Specifically, we define a formal world where hallucination is defined as inconsistencies between a computable LLM and a computable ground truth function. By employing results from learning theory, we show that LLMs cannot learn all the computable functions and will therefore inevitably hallucinate if used as general problem solvers. Since the formal world is a part of the real world which is much more complicated, hallucinations are also inevitable for real world LLMs. Furthermore, for real world LLMs constrained by provable time complexity, we describe the hallucination-prone tasks and empirically validate our claims. Finally, using the formal world framework, we discuss the possible mechanisms and efficacies of existing hallucination mitigators as well as the practical implications on the safe deployment of LLMs.
hub tools
citation-role summary
background 5
citation-polarity summary
roles
background 5representative citing papers
EquiMem calibrates shared memory in multi-agent debate by computing a game-theoretic equilibrium from agent queries and paths, outperforming heuristics and LLM validators across benchmarks while remaining robust to adversarial agents.
Green Shielding introduces CUE criteria and the HCM-Dx benchmark to demonstrate that routine prompt variations systematically alter LLM diagnostic behavior along clinically relevant dimensions, producing Pareto-like tradeoffs in plausibility versus coverage.
DEJA uses evolutionary optimization guided by an LLM-based Answer Utility Score to induce soft-failure responses in RAG systems, achieving over 79% soft attack success rate with under 15% hard failures and high stealth across models and datasets.
A system combining VLM landmark instructions with real-time corrective spatial audio reduces route deviations in a small user study compared to VLM-only and Google Maps audio baselines.
Domain-specialized small language models enable deterministic atomic-resolution scanning probe microscopy control with 99.3% command accuracy, lower computational cost, and better domain performance than larger general models.
VidHal is a new benchmark that evaluates VLLM temporal hallucinations through a caption ordering task on videos with varying hallucination levels.
HistoRAG embeds historiographical principles into RAG via temporal windowing, decoupled retrieval, and contestable LLM relevance judgments, evaluated on 102k Der Spiegel articles from 1950-1979.
RISC reformulates self-consistency answer selection as a ranking task solved by a lightweight LambdaRank model with five hand-designed features, yielding better accuracy-efficiency trade-offs than majority voting on QA benchmarks.
Token entropy distributions fingerprint hallucinations in generative models, enabling the Calibrated Entropy Score (CES) for single-pass black-box detection with calibration guarantees via a novel DKW inequality.
Intent Signal Theory formalizes four distinct intent-related objects in human-AI interaction, introduces a theorem on irreversible private intent loss, and reports supporting patterns from studies across LLMs, languages, and tasks.
Factual recall quality in LLMs follows a sigmoid scaling law in the log-linear combination of model parameter count and topic frequency in training data, explaining 60% of variance across models and up to 94% within families.
Dimension-level evaluation reveals that 25-58% of LLM outputs with perfect holistic scores still show measurable intent deficits across languages and domains.
A dual-purpose benchmark supplies two text-derived knowledge graphs and one expert reference graph on the same biomedical corpus to jointly measure construction method quality and GNN robustness via semi-supervised node classification.
An AI model produced a new formula for a central element of U_q(so_12) at the quality level of advanced undergraduate research, along with faster computation via SageMath, prompting changes in mentorship practices.
LLMs need metacognition to align expressed uncertainty with their actual knowledge boundaries, moving beyond knowledge expansion to reduce confident errors.
An adaptive test-time framework uses a warm-up phase on the test set to build evolving in-context examples, then concentrates compute on unresolved queries to outperform static baselines on math, coding, and reasoning tasks with lower total inference cost.
FineSteer decomposes inference-time steering into Subspace-guided Conditional Steering and Mixture-of-Steering-Experts to deliver stronger control over LLM behaviors with less utility loss than prior methods.
An accurate and trusted AI system cannot achieve human-level reasoning because there exist tasks easily solvable by humans but not by the system.
CogDriver-Agent with sparse temporal memory and spatiotemporal distillation on CogDriver-Data achieves 22% higher closed-loop Driving Score on Bench2Drive and 21% lower mean L2 error on nuScenes.
Introduces a Bayesian framework viewing LLM prompts as textual parameters and proposes MHLP, a novel MCMC algorithm using LLM proposals, to perform inference and improve accuracy plus uncertainty quantification on benchmarks.
Introduces six-dimension trustworthiness definition and attention-based A-Trust score with a TMS to improve LLM-MAS robustness against malicious or unreliable messages.
Hallucinations are inevitable on an infinite set of inputs but can be made statistically negligible with sufficient training data quality and quantity.
A curated set of one billion personas enables scalable, diverse synthetic data generation for LLM training across reasoning, instructions, knowledge, NPCs, and tools.
citing papers explorer
-
Limitations on Accurate, Trusted, Human-level Reasoning
An accurate and trusted AI system cannot achieve human-level reasoning because there exist tasks easily solvable by humans but not by the system.
-
CogDriver: Integrating Cognitive Inertia for Temporally Coherent Planning in Autonomous Driving
CogDriver-Agent with sparse temporal memory and spatiotemporal distillation on CogDriver-Data achieves 22% higher closed-loop Driving Score on Bench2Drive and 21% lower mean L2 error on nuScenes.
-
Textual Bayes: Quantifying Prompt Uncertainty in LLM-Based Systems
Introduces a Bayesian framework viewing LLM prompts as textual parameters and proposes MHLP, a novel MCMC algorithm using LLM proposals, to perform inference and improve accuracy plus uncertainty quantification on benchmarks.
-
To trust or not to trust: Attention-based Trust Management for LLM Multi-Agent Systems
Introduces six-dimension trustworthiness definition and attention-based A-Trust score with a TMS to improve LLM-MAS robustness against malicious or unreliable messages.
-
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.
-
Mitigating hallucinations and omissions in LLMs for invertible problems: An application to hardware logic design automation
Using LLMs to encode logic condition tables into HDL code and decode back to tables mitigates hallucinations in hardware design automation.
-
Multi-agent Self-triage System with Medical Flowcharts
A multi-agent conversational system using AMA flowcharts achieves 95.29% top-3 retrieval accuracy and 99.10% navigation accuracy on large synthetic medical conversation datasets.
-
Hierarchical Memorization in Large Language Models: Evidence from Citation Generation
LLMs memorize citations hierarchically: titles and first authors are recalled at lower redundancy levels than venues or years, with accuracy scaling log-linearly and saturating near verbatim reproduction above roughly 1200 citations.
-
TSGuard: Automated User-Centric Incident Diagnosis for AI Workloads in the Cloud
TSGuard builds domain knowledge bases offline from historical incidents and applies online multi-agent structured reasoning to diagnose AI workload failures, delivering 19.8% higher accuracy and 63.4% lower verification time than baselines on Azure production data.
-
Ask Patients with Patience: Enabling LLMs for Human-Centric Medical Dialogue with Grounded Reasoning
APP is a multi-turn LLM framework for medical dialogue that combines empathetic questioning, Bayesian active learning, and guideline-based reasoning, outperforming baselines on a new simulated-patient benchmark in accuracy, uncertainty reduction, and user experience.
-
Bridging Brains and Machines: A Unified Frontier in Neuroscience, Artificial Intelligence, and Neuromorphic Systems
A position and survey paper that identifies convergence between neuroscience, AGI, and neuromorphic computing and outlines four key integration challenges.
-
Position: Multimodal Large Language Models Can Significantly Advance Scientific Reasoning
Position paper claims multimodal LLMs can significantly advance scientific reasoning and proposes a four-stage roadmap plus challenges and suggestions.