The field of natural language processing is witnessing significant developments in addressing the issue of hallucinations in large language models (LLMs). Hallucinations refer to the generation of non-factual or inaccurate information by LLMs, which can undermine their reliability and trustworthiness. Recent research has focused on developing innovative methods to mitigate hallucinations, including the use of retrieval-augmented generation, context selection, and uncertainty estimation. These approaches aim to improve the accuracy and faithfulness of LLMs by grounding their responses in external knowledge and reducing the reliance on internal parametric knowledge. Noteworthy papers in this regard include Influence Guided Context Selection for Effective Retrieval-Augmented Generation, which introduces a novel metric for context quality assessment, and HalluGuard: Evidence-Grounded Small Reasoning Models to Mitigate Hallucinations in Retrieval-Augmented Generation, which presents a small reasoning model for classifying document-claim pairs as grounded or hallucinated. Overall, the field is moving towards developing more robust and reliable LLMs that can be trusted for real-world applications.
Advances in Mitigating Hallucinations in Large Language Models
Sources
Influence Guided Context Selection for Effective Retrieval-Augmented Generation
Context Is What You Need: The Maximum Effective Context Window for Real World Limits of LLMs
Beyond RAG vs. Long-Context: Learning Distraction-Aware Retrieval for Efficient Knowledge Grounding
LUMINA: Detecting Hallucinations in RAG System with Context-Knowledge Signals
Black-Box Hallucination Detection via Consistency Under the Uncertain Expression
Fine-Grained Detection of Context-Grounded Hallucinations Using LLMs
Multi-level Diagnosis and Evaluation for Robust Tabular Feature Engineering with Large Language Models
LLM-Assisted News Discovery in High-Volume Information Streams: A Case Study
On-Premise AI for the Newsroom: Evaluating Small Language Models for Investigative Document Search
Not Wrong, But Untrue: LLM Overconfidence in Document-Based Queries
ReFACT: A Benchmark for Scientific Confabulation Detection with Positional Error Annotations
UncertainGen: Uncertainty-Aware Representations of DNA Sequences for Metagenomic Binning
SafePassage: High-Fidelity Information Extraction with Black Box LLMs
Copy-Paste to Mitigate Large Language Model Hallucinations
HalluGuard: Evidence-Grounded Small Reasoning Models to Mitigate Hallucinations in Retrieval-Augmented Generation
How can AI agents support journalists' work? An experiment with designing an LLM-driven intelligent reporting system
Efficient Uncertainty Estimation for LLM-based Entity Linking in Tabular Data