Stating false facts confidently

Claims

• mechanismsingle paper↑1↓0
  A ~2M-parameter classifier head that reads hidden states already computed during generation (tapping shallow, middle and deep layers of the generating model) detects response-level hallucination at least as well as specialized hidden-state hallucination detectors, and its detection quality rises with the scale of the base model it is attached to.
• mechanismsingle paper↑1↓0
  In 7-8B open-weight instruct models (Llama-3.1, Mistral-v0.3, Qwen2.5), the drop in a token's probability when a contiguous mid-layer MLP span is zero-ablated is near zero for function words, elevated for knowledge-bearing tokens, and abnormally spiked for hallucinated tokens, so this internal attribution signal — approximable by a small probe from one forward pass — can be used in beam scoring to raise factuality on open-ended and long-form generation, while leaving strict short-answer EM/F1 unchanged or worse.
• observationsingle paper↑1↓0
  In 7-8B open-weight instruct models (Mistral-7B, Qwen2.5-7B, Llama-3-8B) answering closed-book factual questions, a small set of feed-forward neurons (top 1% or fewer) selected on one dataset by class-conditional separation of their residual-space contribution at the final prompt token transfers to other factual QA datasets, where probes on those neuron activations detect hallucination about as well as probes on full residual-stream or MLP-output states.
• mechanismsingle paper↑1↓0
  Accuracy on a fact scales with how many pretraining documents mention it, so long-tail facts are systematically unreliable.
• observationsingle paper↑1↓0
  On adversarial video hallucination pairs scored only when a method both accepts the factual statement and rejects the counterfactual one, purpose-built video hallucination detectors (embedding-similarity, learned-entailment, atomic-fact and attention-based) top out around 34.63% overall accuracy, well below a general frontier video-language model prompted to judge the same items (about 83.63%).
• mechanismsingle paper↑1↓0
  For agentic tool calls where a parameter is fabricated rather than grounded in the dialogue, sampling-based uncertainty detectors (token entropy, semantic entropy, SEP) fail at near-chance because the model fabricates plausible parameters with high confidence, while the model's hidden states at the moment the parameter is generated do carry a separable grounding signal.
• observationsingle paper↑1↓0
  When asked to optimize code that is already at its performance ceiling, frontier LLMs (GPT, Claude, Gemini families) never say it is already optimal under a plain "optimize this code for speed" prompt — they always emit an edit with a confident, unmeasured speedup claim; adding an explicit instruction to abstain unless >90% confident of an improvement produces abstention on a minority of such cases while leaving the edit rate on genuinely sub-optimal code unchanged.
• observationsingle paper↑1↓0
  When asked to supply a missing biomedical reference, models that retrieve a real paper often still misstate its journal, year or authors, so verifiable-reference rate overstates usable accuracy: two Claude models with near-identical verifiable rates (77.6% and 76.6%) were correct in every bibliographic field in 54.6% and 19.9% of responses.
• mechanismsingle paper↑1↓0
  For binary visual question answering with open LVLMs (InstructBLIP-7B, Qwen2.5-VL, InternVL-4B), a hallucinated answer can be held with high response consistency, so a single consistency/uncertainty score misses it; separating consistency into four probe groups (image vs. text perturbation crossed with affirmative vs. negated statement) and classifying on the joint within- and between-group agreement pattern detects hallucination better than any single-score black-box or white-box baseline.
• mechanismsingle paper↑1↓0
  For black-box hallucination detection using only sampled responses and token log-probabilities, semantic-entropy signals go to zero whenever the samples collapse into one semantic cluster — which happens for a large share of hallucinated queries, up to nearly all of them on tasks with confusable entities and figures — while token-level uncertainty still separates many of those cases, so the two signals fail on different queries and combining them detects more than either alone.
• mechanismsingle paper↑1↓0
  For claim-level hallucination detection on long-document QA, estimating a claim's reliability from how central it is in a graph of sampled model responses cannot separate true claims from confidently repeated false ones, because both get high agreement; adding trusted source documents as anchor nodes with higher-weight edges shifts the basis from internal consensus to external grounding and yields a Grounded pool with few hallucinations (97.6% precision on QASPER), at the cost of conservatively leaving correct-but-novel claims on the boundary.
• mechanismsingle paper↑1↓0
  In claim-level hallucination detection pipelines over image, video and audio inputs (both comprehension and generation directions), decomposing an output into atomic claims contributes more to detection accuracy than ensembling several verifier models: removing decomposition costs about 6.6-7.9 macro-F1 across six cross-modal tasks while removing multi-expert majority voting costs about 4.5-5.3.
• mechanismsingle paper↑1↓0
  When claim-level uncertainty is estimated from a response-claim graph, anchoring the graph with trusted external priors (edges weighted by entailment from verified source sections) and scoring claims by weighted closeness centrality separates ungrounded claims from repeatedly-generated-but-false ones better than internal-consistency centrality alone, and yields retrieval whose answerability tracks its accuracy instead of the near-100% answer rate GraphRAG shows on the same long documents.
• mechanismreplicated↑1↓0
  For closed-book hallucination detection, training a small supervised classifier over many heterogeneous uncertainty scores beats the best single score, but the gain comes from the diversity of sampling-based black-box signals: ensembles over only single-generation token-probability features add little, while black-box-only ensembles do nearly as well as the full set.
• observationsingle paper↑1↓0
  On code prompts that are unsatisfiable by design, open-weight code and reasoning models refuse based on how suspicious the request looks rather than how deeply impossible it is: they refuse famous theoretical impossibilities far more often than requests for plausible-sounding nonexistent packages, and the stronger models in the set close the gap on theory while barely improving on fabricated ecosystem entities.
• mechanismsingle paper↑1↓0
  Decomposing long answers into atomic facts and checking each with search catches most errors and beats human annotators on cost and agreement.
• mechanismsingle paper↑1↓0
  For detecting hallucinated text spans in vision-language model outputs, span localization and confidence calibration come from different architectures: a fine-tuned generative VLM proposes better span boundaries but scores them poorly, while a discriminative token tagger trained with a direct regression loss on human-consensus confidence gives better-correlated probabilities but conservative spans, so using the generative spans as proposals and the tagger's probabilities as scores beats either alone.
• observationsingle paper↑1↓0
  When extracting structured rules from multimodal enterprise guideline documents, wrapping the extraction in deterministic schema validation, an LLM semantic judge, and a duplicate/contradiction consistency filter over a shared rule store lowers the share of extracted rules with no matching source sentence from about 15.7% to 3.2% relative to a one-pass pipeline using the same parser and models.
• mechanismsingle paper↑1↓0
  For factual paired-candidate QA where the model already holds the needed knowledge, hallucination can be detected by removing the query keyword that most drives the chosen answer and seeing whether the prediction systematically recovers to the alternative — erroneous answers show that recovery while correct answers merely become ambiguous — but the signal weakens when the knowledge is absent and largely fails on multi-step arithmetic reasoning, where errors are distributed across steps rather than bound to one keyword.
• mechanismsingle paper↑1↓0
  For filtering unsupported LLM-generated vulnerability reports in security triage, statistical machine-text detectors are the wrong instrument: they score provenance, not correctness, and human-written security reports are template-driven and low-perplexity, so they trip the same signal that flags machine text.
• mechanismsingle paper↑1↓0
  When fine-tuning a base model on instruction data, removing from the training targets any factual claim the base model cannot consistently recall raises the supported-claim rate of later long-form generations, but the gain comes largely from a more conservative response policy — more refusals and fewer supported claims per answer — not from generating more correct facts at equal coverage.
• observationsingle paper↑1↓0
  General-purpose factuality verifiers (entailment-, consistency-, and RAG-trained) and zero-shot frontier LLM judges perform near chance at spotting unsupported claims in scientific peer reviews that must be checked against the full submitted paper, because they cannot separate paper-unsupported assertions from legitimate evaluative critique; small models fine-tuned on in-domain examples beat them by a wide margin.
• observationsingle paper↑1↓0
  General-purpose factuality verifiers (entailment-, consistency-, and RAG-trained) and zero-shot frontier LLM judges perform near chance at spotting unsupported claims in scientific peer reviews that must be checked against the full submitted paper, because they cannot separate paper-unsupported assertions from legitimate evaluative critique; small models fine-tuned on in-domain examples beat them by a wide margin.
• mechanismsingle paper↑1↓0
  Grounding generation in retrieved documents improves factual accuracy on knowledge-intensive tasks.
• mechanismsingle paper↑1↓0
  When a language model is updated by fine-tuning on a substituted fact, the facts it already answered correctly that are most likely to flip to wrong are those whose answer entity is referenced by many other facts, and updates to such structurally popular facts corrupt connected facts out to five hops in the fact graph — so popularity, not long-tail rarity, predicts collateral corruption among already-correct knowledge.
• mechanismsingle paper↑1↓0
  In large audio-language models answering binary "is sound X present" questions, the yes/no outcome is fixed at the first decoding step and the model's yes-minus-no margin is positive for most clips even with a silent input, so penalizing affirmative tokens at that single step whenever real audio fails to raise the margin above a small threshold increases correct rejection of absent events beyond fixed-strength contrastive decoding, at the cost of lower recall on genuinely present events and lower precision.
• mechanismsingle paper↑1↓0
  In LLM-based ASR, steering decoding toward acoustic evidence only helps if candidates are first restricted to tokens within a small likelihood margin of the greedy token: unconstrained selection by an acoustic compatibility score removes hallucinated outputs but destroys transcription accuracy through unbounded insertions, while the constrained version removes a large share of hallucinations at near-zero WER/CER cost.
• observationsingle paper↑1↓0
  In LLM-based program repair on Defects4J bugs with frontier models (GPT-5, DeepSeek-R1, Claude Sonnet 4.5) prompted zero-shot, passing the developer-written test suite does not mean the patch is grounded in the bug: manual inspection of 812 sampled repairs found ungrounded repairs in 72.7% of cases including test-passing patches, most often from incorrect causal localization (45.9%) rather than a wrong edit style, and the same models frequently misidentify the bug-triggering tests and generate additional tests that miss the triggering condition.
• mechanismsingle paper↑1↓0
  Long-form biographies from strong models contain a substantial share of unsupported atomic facts, rising for less well-known subjects.
• observationsingle paper↑1↓0
  When a model is asked to audit many documents for planted errors in a single large-batch prompt, recall of known contaminants collapses and the model reports confident fabricated findings — invented absurd contaminants that appear in no document — rather than signalling incomplete processing.
• mechanismsingle paper↑1↓0
  Models reproduce common human misconceptions, and larger models were not more truthful on this benchmark.
• mechanismsingle paper↑1↓0
  In multi-agent LLM systems where one agent starts with an erroneous shared belief, broadcasting every agent's full context at every step raises the rate of false assertions above doing no synchronization at all, because the error propagates to agents that were previously correct — and the harm appears only in tasks where one wrong fact cascades across semantically linked dimensions (destination to airport to weather), not where agent contexts are largely orthogonal.
• observationsingle paper↑1↓0
  For multi-hop factual QA with Qwen2.5-7B/Qwen3-8B given a deliberately insufficient retrieved triple subgraph, training a policy with RL to insert bridging triples that let a separate frozen answerer reach the gold answer raises answer accuracy over standard GRPO and over prompting methods like Self-Ask and RECITE, and the improvement carries over to direct chain-of-thought answering with no triples inserted at test time.
• observationsingle paper↑1↓0
  When multiple stochastic runs of several mid-tier LLMs each produce an unranked set of candidate security requirements from the same control templates, pooling the runs recovers far more valid items than any single run but also many hallucinations, and ranking the pooled candidates by how many runs produced them separates valid from hallucinated items well enough to beat every individual run's precision-recall and ROC area.
• observationsingle paper↑1↓0
  For open-ended emotion interpretation by multimodal LLMs, coarse-grained hallucination mitigations designed for object hallucination (VCD, ICD, M3ID) and self-verification (EmotionHallucer) cut unsupported claims in some facets while raising them in others — e.g. lowering audio hallucination but increasing expression and action hallucination — so a single aggregate hallucination rate hides the trade-off.
• mechanismsingle paper↑1↓0
  In an open-ended LLM roleplay game where the character must withhold authored facts, gating generation through an external state tree — one retrieved permitted fact injected into the prompt, plus a separate verifier that decides whether the fact was actually conveyed before updating state — cuts the share of responses containing story-breaking unauthored content (17.8% to 6.3%) and stops premature reveal of the ending entirely, but makes some reveals appear unmotivated because the generator is forced to deliver the retrieved fact on that turn.
• mechanismsingle paper↑1↓0
  In open-weight 7-8B instruct models, a supervised detector built on how hidden states change between adjacent layers separates factually wrong from correct answers better than a probe on any single layer, and shuffling the layer order destroys most of the signal — so the ordering of the depth trajectory, not just the content of one layer, carries the information.
• mechanismsingle paper↑1↓0
  When post-training a speech LLM for ASR with GRPO, adding a reward term from a separately pretrained, frozen CTC acoustic judge that scores each sampled transcription against the audio cuts insertion errors on out-of-distribution meeting speech (22-28% fewer on AMI) relative to a schedule-matched text-only -WER reward, without raising word error rate and without needing the judge at inference.
• mechanismsingle paper↑1↓0
  On questions whose premises are false or internally contradictory, prompting mid-size instruct models (Qwen2.5-7B, LLaMA-3.1-8B, Gemma3-12B) to reason step by step can lower accuracy below plain answering, because the reasoning chain elaborates from the flawed premise instead of challenging it.
• mechanismsingle paper↑1↓0
  For response-level hallucination detection against reference documents, having the model first convert the reference into a relational SQL database and then check response claims by executing SELECT queries raises F1 over zero-shot direct prediction mainly by cutting false positives, and the gain shrinks or vanishes on backbones with weak text-to-SQL ability.
• observationsingle paper↑1↓0
  For response-level hallucination detection in retrieval-augmented settings, having the model first convert the reference documents into a relational SQL database and then check the response against SELECT query results raises precision and macro F1 over zero-shot direct prediction, over atomic-claim decomposition, and over flat JSON key-value extraction, without any task-specific fine-tuning.
• mechanismsingle paper↑1↓0
  In retrieval-augmented answers, splitting an answer into claims, checking each against the retrieved source, and then acting on the flagged claims reduces the share of answers judged to contain unsupported content — but the strategies trade grounding against preservation: deleting unsupported claims reduces unsupported content most while retaining the least original text, and rewriting retains the most while reducing least.
• mechanismreplicated↑2↓0
  Grounding answers in retrieved documents reduces unsupported generation but transfers the trust problem to the retrieval channel: whatever lands in context is treated as reliable, which shows up both as following instructions planted in retrieved content and as deferring to whichever account is better represented when sources disagree.
• observationsingle paper↑1↓0
  When a retrieval store holds both the correct answer and a common misconception for the same question, injecting all retrieved memories raises the hallucination rate above a memory-free baseline for weaker models (53.0% vs 23.0% on TruthfulQA with gemma-4-E4B-it), while a stronger backbone (deepseek-v4-flash) instead shows RAG below its own no-memory rate, so conflict-driven amplification is concentrated in models susceptible to memory injection.
• mechanismsingle paper↑1↓0
  When retrieval for a summarization task pulls definitions by surface term match from a general encyclopedia, ambiguous terms fetch the wrong sense and the model incorporates that unsupported background into its output, lowering claim-level factuality below the no-retrieval baseline even when the prompt tells it to ignore unrelated retrieved content.
• mechanismsingle paper↑1↓0
  In semi-extractive RAG QA where the model is prompted to put quoted evidence in brackets, the bracketed spans frequently do not appear verbatim in the retrieved context (as low as 0.37 verbatim rate on a medical dataset with Mistral-Nemo), and masking logits during bracketed spans to only tokens continuing a verbatim context span raises verbatim fidelity to near 1.0 across Llama-3.1-8B, Mistral-Nemo and Qwen3-4B without lowering ROUGE-L or BERTScore much.
• observationsingle paper↑1↓0
  On single-step retrosynthesis for real industrial target molecules, frontier general LLMs produce chemically infeasible top-1 routes (unresolved selectivity or no mechanistic basis) more often than specialized retrosynthesis models, even though their reference-matching recall is comparable, so disconnection fluency does not carry selectivity awareness.
• observationsingle paper↑1↓0
  For small open-weight models (1.5-4B) generating answers in a closed-domain RAG setup, hallucinated generations show a broken hidden-state proximity graph between the retrieved context entity and the generated entity — connectivity below 20% and path lengths above 7 hops, versus near-full connectivity at about 3 hops for grounded answers — giving a training-free detection signal (AUROC 0.89 on Qwen2.5-1.5B).
• observationsingle paper↑1↓0
  For small open-weight Qwen2.5 models (1.5B-3B) doing multi-step math reasoning, features derived from the spread of attention entropy across layers at each generation step predict whether the final answer is wrong better than output-logit confidence features, with the margin larger on harder problems (+0.0093 AUC on GSM8K, +0.0759 on MATH-500 at 3B).
• mechanismsingle paper↑1↓0
  Surveys causes, kinds, and detection of hallucination across data, training, and inference, and finds no complete solution.
• observationsingle paper↑1↓0
  When tool-augmented agents emit calls over an unconstrained raw-JSON surface rather than a schema-enforcing function-calling API, every tested hosted model fabricates tool names or undeclared arguments at rates of roughly 13-57%, and parameter count does not reduce this — an 8B and a 675B model hallucinate at comparable raw-JSON rates.
• observationsingle paper↑1↓0
  For tool-augmented legal agents on multi-step Chinese-law tasks, a correct final answer does not indicate a clean reasoning trajectory: across 18 agent configurations, trajectories that reach the right answer still contain at least one substantive legal hallucination about 68% of the time on average, and that rate barely moves with backbone scale or agentic framework (0.61-0.78), while procedural-process hallucination rates vary widely (0.07-0.85).
• mechanismsingle paper↑1↓0
  When a tool call is forced and returns an unusable payload, tool-augmented agents report the failure honestly almost always if the response envelope declares an error, but assert unsupported values or invent policy reasons in a large share of cases when the envelope says status:ok while withholding the value (redacted, corrupted, stale, empty, malformed, truncated).
• mechanismsingle paper↑1↓0
  For tool-calling agents that fabricate parameters never given by the user, a zero-initialized LoRA adapter run concurrently with the frozen base model and trained only to emit a localized verdict at an appended trigger token detects these ungrounded parameters better than sampling-based entropy methods, passive linear probes, and an equivalent-scale fine-tuned external judge, and naming the specific hallucinated parameter back to the agent recovers more of the task than a generic block.
• mechanismsingle paper↑1↓0
  Training a language model with differential privacy — whether DP-SGD pre-training or DP LoRA fine-tuning — makes it generate more false claims about facts in its training data than a non-private counterpart, and the effect grows as the privacy budget tightens, because per-example gradient clipping plus noise prevents low-frequency facts from being acquired at all while flattening the next-token distribution onto incorrect alternatives.
• mechanismsingle paper↑1↓0
  In whole-repository cross-language code migration (Java to Python), supplying an LLM with AST-derived dependency-graph context and translating parent classes before children lowers the rate of calls to nonexistent APIs relative to vector-similarity retrieval of text chunks, while lexical similarity metrics such as CodeBLEU show no difference between the two.
• observationsingle paper↑1↓0
  In zero-shot LLM program repair on Defects4J bugs, passing the developer-written test suite does not indicate that the patch is grounded in a correct understanding of the bug: manual inspection of sampled repairs found ungrounded patches in about three quarters of cases including ones that passed all available tests, with incorrect causal localization the largest category.

Techniques

• Anchor claim-consistency graphs to trusted priorsretrieval
  Decompose sampled responses and trusted source documents into atomic claims, build a weighted bipartite graph where prior-entailed edges cost least, and label each claim Grounded, Refuted, or Boundary by weighted closeness centrality.
• Bound audit batch size and verify every reported finding agaprocess
  Restrict document-auditing prompts to small batches with content injected directly, and mechanically check each reported finding against the source text before accepting it.
• Check long answers fact by facttooling
  Split a long answer into atomic claims and verify each against search or a corpus.
• Gate disclosure through a structured knowledge tree with retprocess
  Store the authored facts, prerequisites and contradictions in a tree; one model picks the single currently-permitted unrevealed fact, a second writes dialogue constrained to it, and a third checks whether the fact was conveyed before the tr
• Gate extraction through a versioned rule store with schema aprocess
  Every extracted rule is written to a shared versioned store under a stable id, checked against 26 schema constraints, scored on five dimensions by an LLM judge whose minimum dimension decides auto-approve/review/reject, then filtered for du
• Ground answers in retrieved documentsretrieval
  Retrieve relevant documents and generate from them, with the model deciding when retrieval is needed.
• Judge each rubric item against the full agent trajectoryprocess
  Convert a verified gold answer into a checklist whose items are each anchored to a hallucination subclass, then have an LLM judge issue a binary verdict per item against the agent's whole trajectory (reasoning, memory, tool observations) ra
• LoRA 'Latent Critic' that verbalizes which parameter is ungrtraining
  A zero-initialized LoRA adapter runs concurrently with the frozen base model during tool-call generation, trained with loss only on a trigger token and label, so it restructures the residual stream and then emits a localized verdict like 'u
• Mask decoding to verbatim context spans inside quotesdecoding
  When a quote-opening token is generated, restrict the next-token logits (via a suffix tree over tokenised retrieved contexts) to tokens that extend the current span into a continuous verbatim substring of some retrieved passage, plus the cl
• Probe consistency on four axes, not one scoreprompting
  Convert a binary question into an affirmative and a contradictory statement, re-ask each under semantically equivalent image transformations and paraphrases to form four consistency groups, and train a lightweight classifier on 14 within-gr
• Rank feed-forward neurons by class-conditional residual contarchitecture
  Score each MLP neuron at the final prompt token by the separation between hallucinated and truthful mean pre-to-post activation change scaled by its output-vector norm, then train a small classifier on the top-ranked neurons' features.
• Rank pooled multi-run outputs by cross-run vote countprocess
  Treat each LLM run as a black-box retrieval system casting a binary vote per candidate requirement, sum the votes (optionally weighted by each model family's estimated true-positive/false-positive log-odds), and hand reviewers the resulting
• Reject tool calls that do not resolve to a registered tool atooling
  Before any permission gate, check each emitted call's name against the closed-world tool registry and type-check its arguments against the declared signature, rejecting anything unresolved.
• Rerank candidate routes with an executable chemical verifierprocess
  Take any backbone's top-10 candidate disconnections and reorder them by rule-based checks for functional-group competition, mechanistic plausibility and literature precedent, promoting unflagged routes above high-confidence flagged ones.
• Reward acoustic fidelity with a frozen judge during RLtraining
  Augments the text-space -WER GRPO reward with a group-normalized score from a permanently frozen, non-autoregressive CTC model that measures how well each sampled transcription aligns to the audio, used at training time only.
• Score detectors on paired factual/counterfactual decisionsprocess
  Evaluate a hallucination detector by requiring it to accept the factual answer and reject the minimally edited counterfactual for the same video, reporting the joint success rate rather than one-sided accuracy.
• Score intermediate repair artifacts, not just the patchprocess
  Require the repair model to also produce checkable intermediate artifacts (which tests trigger the bug, which lines execute before and after the patch, new bug-exposing tests) and grade those against execution evidence as a reliability sign
• Score reasoning failure by cross-layer attention-entropy distooling
  At each generation step, average attention entropy per layer, take the standard deviation across layers, and aggregate these over the sequence as features for a logistic-regression hallucination classifier.
• Suppress hallucinated logit directions and re-anchor on veridecoding
  At each sentence, an evaluator localizes hallucinated claim tokens into a hallucination memory used to build a vocabulary-level suppression direction applied to the logits, while the lowest-hallucination sentence so far is kept in an anchor
• Taxonomy-driven hallucination injection for review verificatprocess
  Taxonomy-driven hallucination injection for review verification benchmarks, as introduced by the paper it was drafted from.
• Verify responses against an LLM-built SQL database of the sotooling
  Prompt the model to incrementally build a relational SQL database from the reference passages, generate SELECT queries for the claims in the response, and reconcile its initial neural hallucination judgement with the executed query results.

Sources

Every paper, post or observation behind this capability. Ones cited by a claim show the claim and its stance; ones filed directly are about the capability but have no claim drafted from them yet. Open one to add support or contest it.

• Enforcing Narrative Reliability and Epistemic Pacing in LLM-Driven Detective Games via Structured Knowledge Trees2026-09-19
  ↑ supports a claim here
• An Interpretable Memory Decision Controller for LLM Agents Based on Three-Signal Complementarity: Decoupling Confidence and Consistency2026-09-18
  ↑ supports a claim here
• Attention Dispersion as a Diagnostic Signal for Hallucination in Large Language Models2026-09-16
  ↑ supports a claim here
• Closed-World Resolution Against Tool Hallucination in LLM Agents2026-09-16
  ↑ supports a claim here
• SAGE: Governed Artifact Generation from Enterprise Guidelines2026-09-15
  ↑ supports a claim here
• SAVOR: Self-Aware Visual Grounding via Confidence-Calibrated Reinforcement Learning for Multimodal Hallucination Mitigation2026-09-15
  about this capability · no claim drafted yet
• Semantic-Spatial Agreement Verification for Mitigating Object Hallucination in Multimodal Large Language Models2026-09-15
  about this capability · no claim drafted yet
• What Do Hallucinations Reveal About Multimodal Reasoning? Diagnosing Visual Grounding Failures via Contrastive Decoding Probes2026-09-15
  about this capability · no claim drafted yet
• AgenTeeth: A Model-Agnostic Framework for Suppressing Hallucination in Frozen Vision-Language Models on Dental X-Rays via Tool Evidence Injection2026-09-15
  about this capability · no claim drafted yet
• Look Before You Leap: Factual Decoding with Internal Attribution Signals2026-09-14
  ↑ supports a claim here
• Biomedical Reference Generation Remains Unreliable across 26 Large Language Models2026-09-14
  ↑ supports a claim here
• When the Wrong Key Wins: Understanding and Detecting Hallucinations in LLMs2026-09-14
  ↑ supports a claim here
• NeuroActiSep: Detecting Factual Hallucinations from Feed-Forward Neurons in a Single Pass2026-09-13
  ↑ supports a claim here
• Efficiency Hallucination: Formalizing and Measuring Behavioral Calibration in LLM-Based Code Optimization2026-09-13
  ↑ supports a claim here
• Grounded in Sound: Reinforcement Learning with a Frozen Acoustic Judge to Curb ASR Insertion Hallucinations2026-09-13
  ↑ supports a claim here
• Fabrication After Tool Failure: Tool-Augmented Agents Assert Values Their Tools Did Not Return2026-09-13
  ↑ supports a claim here
• E2A-Bench: Benchmarking Evidence-to-Action Reliability in Financial Chart Reasoning2026-09-13
  about this capability · no claim drafted yet
• Beyond Vector Similarity: Hierarchical Context-Aware Graph RAG vs Standard RAG in Enterprise Code Migration2026-09-11
  ↑ supports a claim here
• Building a Production Greek-English Speech Recognizer2026-09-11
  about this capability · no claim drafted yet
• OmniHallu: Unified Hallucination Detection for Cross-Modal Comprehension and Generation in Multimodal Large Language Models2026-09-10
  ↑ supports a claim here
• Multi-Faceted Evaluation and Mitigation of Emotion Hallucinations in MLLMs2026-09-10
  ↑ supports a claim here
• Domain-Specific Hallucination Detection in Large Language Models2026-09-10
  about this capability · no claim drafted yet
• NovGauge: A Fine-Grained Benchmark for Diagnosing LLMs' Capability in Paper Novelty Assessment2026-09-10
  about this capability · no claim drafted yet
• The Cost of Compression: A Rate-Distortion Limit on Factual Hallucination2026-09-10
  about this capability · no claim drafted yet
• Can We Trust Video Hallucination Detectors? VidHalLoc for Evaluating the Evaluators2026-09-09
  ↑ supports a claim here
• When Auditors Fabricate: Batch-Size Degradation and Confident Hallucination in LLM Detection of Planted Document Contamination2026-09-09
  ↑ supports a claim here
• Ensembling LLMs for AI-Augmented Cybersecurity Software Requirements Generation2026-09-09
  ↑ supports a claim here
• Guaranteeing Faithful Evidence Extraction in Speculative Retrieval-Augmented Generation2026-09-09
  ↑ supports a claim here
• LexAgentHallu: A Hierarchical Benchmark for Profiling Hallucinations in Legal Agents2026-09-09
  ↑ supports a claim here
• Popular Knowledge Propagates More Errors in LLM Knowledge Updating2026-09-08
  ↑ supports a claim here
• TAD: Token-Adaptive Contrastive Decoding with Confidence-Guided Gating for Hallucination Mitigation in Large Audio-Language Models2026-09-07
  ↑ supports a claim here
• MLLMs Hallucinate when Information Distribution Drifts in Synergy Heads2026-09-05
  about this capability · no claim drafted yet
• Better Understanding, Better Fixes? A Study of Hallucination in LLM-based Automated Program Repair2026-09-04
  ↑ supports a claim here · ↑ supports a claim here
• Leveraging Low-Level Symbolic Competences for Unsupervised Grounding in Hallucination Detection2026-09-04
  ↑ supports a claim here · ↑ supports a claim here
• GRACE: Graph-Grounded Reflective Agent Copilot Engine for Expert-in-the-Loop Knowledge Expansion2026-09-03
  ↑ supports a claim here · ↑ supports a claim here
• Refusing the Impossible: A Taxonomy and Benchmark for Code Hallucination in Large Language Models2026-09-03
  ↑ supports a claim here
• HalluPeer: A Taxonomy-driven Benchmark for Detecting Hallucinations in Scientific Peer Reviews2026-09-03
  ↑ supports a claim here · ↑ supports a claim here
• From Tokens to Semantics: Leveraging Complementary Signals for Hallucination Detection in Black-Box LLMs2026-09-02
  ↑ supports a claim here
• Improving Health Literacy through Lay Summarization of Radiological Reports: An Evaluation of BioNER and Retrieval-Augmented Generation2026-09-02
  ↑ supports a claim here
• RVSD: Retrieval Vision Sparse Decoding for Mitigating Visual Hallucinations in Large Vision-Language Models2026-09-02
  about this capability · no claim drafted yet
• Does Playing it Safe Count as Faithfulness? Reassessing LVLM Hallucination Mitigation Methods2026-09-01
  about this capability · no claim drafted yet
• Reliability Challenges in Diffusion Vision-Language Models2026-09-01
  about this capability · no claim drafted yet
• SingProbe Technical Report2026-08-31
  ↑ supports a claim here
• Stick to What You Know: A Study of Knowledge-Aligned Supervised Fine-Tuning2026-08-31
  ↑ supports a claim here
• Likelihood-Constrained Acoustic Reranking for Training-Free Hallucination Mitigation in LLM-Based ASR2026-08-31
  ↑ supports a claim here
• The Privacy-Hallucination Tradeoff in Differentially Private Language Models2026-08-31
  ↑ supports a claim here
• Fine-Grained Multi Image Object Hallucination Benchmark2026-08-31
  about this capability · no claim drafted yet
• Beyond Language Priors: Diagnosing and Fixing Visual-Origin Hallucinations in Multimodal LLM2026-08-31
  about this capability · no claim drafted yet
• VisER: Visual Evidence and Reliance for Object Hallucination Detection in LVLMs2026-08-31
  about this capability · no claim drafted yet
• SpanCalib-VLM: Calibrated Hallucination Span Detection in Vision-Language Models2026-08-30
  ↑ supports a claim here
• Hallucination Mitigation for Large Vision-Language Models via Implicit Feature Stabilization2026-08-30
  about this capability · no claim drafted yet
• Detecting and Repairing Hallucinations in Retrieval-Augmented Generation2026-08-29
  ↑ supports a claim here
• EviAnchor: Mitigating Hallucinations in Large Vision-Language Models via Regional Visual Evidence Compensation2026-08-29
  about this capability · no claim drafted yet
• Dynamic Alignment Compensation for Hallucination Mitigation in Large Vision-Language Models2026-08-28
  about this capability · no claim drafted yet
• Prediction of Prediction (PoP): Inter-Layer Activation Fusion for Single-Pass Hallucination Detection in Large Language Models2026-08-27
  ↑ supports a claim here
• AI Slop and Hallucinations in Vulnerability Assessment: A Survey on Reasoning Failures and Trustworthy Mitigation2026-08-26
  ↑ supports a claim here
• From Passive Response to Proactive Correction: Enhancing LLM Robustness Against Input Fact Perturbations2026-08-26
  ↑ supports a claim here
• Overview of SHROOM-Visions 2026: A Shared Task on Hallucination Detection in Large Vision-Language Models2026-08-26
  about this capability · no claim drafted yet
• When Do Supervised UQ Ensembles Improve LLM Hallucination Detection? A Robustness Study2026-08-25
  ↑ supports a claim here
• Lost in Speech: Trilingual Spoken Hallucination Detection Across Audio and Transcripts2026-08-25
  about this capability · no claim drafted yet
• Targeting the Attention Heads Behind Object Hallucination in LLaVA2026-08-25
  about this capability · no claim drafted yet
• Credal Large Language Models for Semantic Commitment under Uncertainty2026-08-24
  about this capability · no claim drafted yet
• From Storage to Access: Verifiable Activation of Parametric Knowledge in LLMs via Explicit Priming and Implicit Reasoning2026-08-19
  ↑ supports a claim here
• ReWEIGH the Evidence: Calibrating Token-Level Ordinal Visual Evidence to Mitigate Hallucinations in Large Vision-Language Models2026-08-19
  about this capability · no claim drafted yet
• Do Large Language Models Play Six Degrees of Separation? Measuring Topological Compression in Long-Context Manifolds2026-08-18
  ↑ supports a claim here
• Counterfactual Anatomy-guided Spatial-Temporal Decoding for Annotation-Free Hallucination Mitigation in Medical VLMs2026-08-18
  about this capability · no claim drafted yet
• Actionable Hallucination Detection: Translating Latent Uncertainty into Agentic Critique2026-08-11
  ↑ supports a claim here · ↑ supports a claim here
• Test-Time Hallucination Control in Large Vision-Language Models2026-08-11
  about this capability · no claim drafted yet
• Beyond Global Editing: Per-Instance Disentangled Subspaces for Training-Free Hallucination Mitigation in LVLMs2026-08-10
  about this capability · no claim drafted yet
• Wiener Representation Filtering for VLM Hallucination Suppression2026-08-08
  about this capability · no claim drafted yet
• Same Attention, Different Truths: Put Logit-Lens over Visual Attention to Detect and Mitigate LVLM Object Hallucination2026-08-07
  about this capability · no claim drafted yet
• TruthLens: Object Hallucination Detection via Self-Evaluating Truthfulness Scores in LVLMs2026-08-06
  about this capability · no claim drafted yet
• UHP Detection: LVLMs have their Unique Hallucination Pattern in the Consistency Space2026-08-04
  ↑ supports a claim here
• SureRoute: Toward a Hallucination-Free Self-Improving Platform for Retrosynthesis2026-08-04
  ↑ supports a claim here
• Hallucination as Context Drift: Synchronization Protocols for Multi-Agent LLM Systems2026-06-19
  ↑ supports a claim here
• Large Language Models Struggle to Learn Long-Tail KnowledgeThis source: Large Language Models Struggle to Learn Long-Tail KnowledgeHow much the field cites it — very heavily cited in the last 12 months203 in the last 12 months · 730 totalpublished 2022checked 2026-09-042022
  ↑ supports a claim here
• Long-form factuality in large language modelsThis source: Long-form factuality in large language modelsHow much the field cites it — heavily cited in the last 12 months75 in the last 12 months · 167 totalpublished 2024checked 2026-09-042024
  ↑ supports a claim here
• Retrieval-Augmented Generation for Knowledge-Intensive NLP TasksThis source: Retrieval-Augmented Generation for Knowledge-Intensive NLP TasksHow much the field cites it — very heavily cited in the last 12 months1000+ in the last 12 months · 17840 totalpublished 2020checked 2026-09-04count capped at 1000 by the fetch2020
  ↑ supports a claim here
• FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text GenerationThis source: FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long…How much the field cites it — very heavily cited in the last 12 months697 in the last 12 months · 1643 totalpublished 2023checked 2026-09-04click for Semantic Scholar2023
  ↑ supports a claim here
• TruthfulQA: Measuring How Models Mimic Human FalsehoodsThis source: TruthfulQA: Measuring How Models Mimic Human FalsehoodsHow much the field cites it — very heavily cited in the last 12 months1000+ in the last 12 months · 3851 totalpublished 2021checked 2026-09-04count capped at 1000 by the fetch2021
  ↑ supports a claim here
• Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt InjectionThis source: Not what you've signed up for: Compromising Real-World LLM-Integrated …How much the field cites it — very heavily cited in the last 12 months1000+ in the last 12 months · 1870 totalpublished 2023checked 2026-09-04count capped at 1000 by the fetch2023
  ↑ supports a claim here
• Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge ConflictsThis source: Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large …How much the field cites it — heavily cited in the last 12 months148 in the last 12 months · 382 totalpublished 2023checked 2026-09-042023
  ↑ supports a claim here
• A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open QuestionsThis source: A Survey on Hallucination in Large Language Models: Principles, Taxono…How much the field cites it — very heavily cited in the last 12 months1000+ in the last 12 months · 3664 totalpublished 2023checked 2026-09-04count capped at 1000 by the fetch2023
  ↑ supports a claim here