The field of natural language processing is moving towards addressing the critical challenge of hallucination in large language models. Hallucination refers to the generation of content that is not faithful to the input or real-world facts. Recent research has focused on developing effective methods for detecting and mitigating hallucinations, including the use of reinforcement learning, entity hallucination indices, and retrieval-augmented generation. These approaches have shown promising results in reducing hallucinations in language models and improving their overall reliability. Noteworthy papers in this area include 'A Survey of Multimodal Hallucination Evaluation and Detection' and 'Theoretical Foundations and Mitigation of Hallucination in Large Language Models', which provide comprehensive overviews of hallucination evaluation benchmarks and detection methods, as well as theoretical analyses and mitigation strategies. Another notable paper, 'First Hallucination Tokens Are Different from Conditional Ones', analyzes the variation of hallucination signals within token sequences and provides insights into token-level hallucination detection.
Hallucination Detection and Mitigation in Large Language Models
Sources
A Survey of Multimodal Hallucination Evaluation and Detection
Advancing Mental Disorder Detection: A Comparative Evaluation of Transformer and LSTM Architectures on Social Media
IFD: A Large-Scale Benchmark for Insider Filing Violation Detection
Enhancing Hallucination Detection via Future Context
First Hallucination Tokens Are Different from Conditional Ones
FRED: Financial Retrieval-Enhanced Detection and Editing of Hallucinations in Language Models
Multi-Amateur Contrastive Decoding for Text Generation
Understanding Public Perception of Crime in Bangladesh: A Transformer-Based Approach with Explainability
Bangla BERT for Hyperpartisan News Detection: A Semi-Supervised and Explainable AI Approach
Investigating Hallucination in Conversations for Low Resource Languages
Reducing Hallucinations in Summarization via Reinforcement Learning with Entity Hallucination Index
A Language Model-Driven Semi-Supervised Ensemble Framework for Illicit Market Detection Across Deep/Dark Web and Social Platforms
Theoretical Foundations and Mitigation of Hallucination in Large Language Models
LLM-Assisted Cheating Detection in Korean Language via Keystrokes
A Single Direction of Truth: An Observer Model's Linear Residual Probe Exposes and Steers Contextual Hallucinations
T-Detect: Tail-Aware Statistical Normalization for Robust Detection of Adversarial Machine-Generated Text