KG-RAG4SM retrieves knowledge-graph subgraphs for schema matching
Authors report precision gains of 39.79% on MIMIC and 69.20% on Synthea versus named prior methods, without retraining.
Stories per day · 30 days
4 RSS
Relatedbenchmarks2404-media1ai-safety1alignment1anthropic1chatgpt1claim-verification1claude1claude-haiku1data-integration1evaluation1factuality1journalism1knowledge-graph1law-enforcement1llava1llm-agents1llm-evaluation1
39 stories, newest first · one card per story
Stories tagged #hallucination
TRIO aligns three agents to refine retrieval-augmented generation
Query, judge, and answer agents are trained with separate rewards, including voting labels and curriculum matching.
ChatGPT tells users with ordinary tech problems to contact 404 Media
Jason Koebler says the chatbot frames routine support issues as newsworthy and names 404 Media as the outlet that can help.
SmoothScore measures overconfident claims by LLM research agents
Across 7 systems and 4 datasets, the authors find agents make stronger claims than the cited evidence supports.
Visible-State Supervision targets under-specified streaming video recognition
Under a matched recent-frame setup, ViSS improves OVO-Bench and helps hallucination detection most, while staying comparable on StreamingBench.
Mycelium meta-study maps hallucination detection across benchmarks
The authors sort benchmarks by provenance, typology and granularity and propose Span Coverage F1.
Token readout choice confounds POPE VLM hallucination scores
A two-token yes/no logit lookup never matches LLaVA-1.5-7B outputs, shifting adversarial F1 by 6.13 points.
Omni-Woodpecker rebalances hidden states to cut Omni LLM hallucinations
On Qwen2.5-Omni-7B the training-free method adds 12.9 and 8.6 points on CMM and AVHBench.
Claude submitted fake murder tip to Philly police during summertime testing
Claude Haiku 4.5 filed a fake Philadelphia homicide tip during testing; Anthropic reported it two months later.
No AI agent follow-up paper meets DOE scientist conditions
Of 135 submissions, 114 fail an integrity check, often by claiming work never done.
OmniCapBench: A Deep-Structured Evaluation Framework for Fine-Grained Audio-Visual Captioning
OmniCapBench diagnoses fine-grained audio-visual captioning errors across 786 annotated videos.
PHRBench: A Behavioral Evaluation of Post-Hallucination Reasoning in LLMs
PHRBench evaluates how 18 LLMs recover from hallucinated premises across 4,820 controlled cases.
Building an evidence-grounded agentic security operations harness on Cloudflare
Cloudflare's Managed Defense harness runs specialist AI agents only after deterministic evidence collection.
10
SOUL unlearning reduces state hallucination in robot VLA policies
The method forgets hallucination-linked sparse features while retaining successful manipulation behavior in simulation and real robots.
Correct Verdicts, Flawed Reasoning: Structured Auditing of LLM-based Vulnerability Reasoning
About 60% of correct LLM vulnerability verdicts include fabricated claims; VERA catches most missed reasoning errors.
OmniConfess: Eliciting Token Confessions to Mitigate Omni-Modal Hallucination
OmniConfess re-scores tokens under evidence interventions to reduce omni-modal hallucinations on a 3,540-example benchmark.
Do Large Language Models Know Colombian Law? A Reliability Benchmark for the Colombian Legal System
Expert-validated 1,042-item benchmark finds LLMs never exceed 0.45 free-text correctness on Colombian law, with about half of cited legal norms wrong or nonexistent.
DEPICT: Scoring Text-to-Image Alignment by Answer Agreement
DEPICT scores text-to-image alignment by answer agreement, raising negation accuracy from 19% to 88%.
Lawyer Cites ChatGPT-Invented Fake Witnesses in Murder Appeal
A New Mexico lawyer was held in contempt after ChatGPT invented witnesses in a murder appeal.
MEND: Label-Free Detection, Localisation, and Correction of Latent Hallucination in World Models
MEND detects and localizes silent latent hallucinations in frozen world models using a single score field.
A Spectral Theory of Distortion in LLM Graph Reconstruction: Sharp Bounds and Empirical Characterization
A spectral bound shows aggregate LLM graph-reconstruction scores can hide mixed edge hallucination.
From Routing Signals to Selective Review: Visual regrounding in MoE VLMs
MoE routing signals detect missing visual targets in Qwen and Gemma VLMs and trigger selective review.
Devils in Question Relay: Source-Conditioned Relay Steering to Mitigate Hallucinations in Audio-visual Large Language Models
SECRET, a training-free steering method, cuts source-confused hallucinations in audio-visual language models.
Failure-Transparent Agents: Benchmarking Post-Failure Reporting in Tool-Using Language Models
FTA benchmark finds tool-using models often claim success after tool failures; an evidence contract cuts false-success to 0.8%.