The field of large language models (LLMs) is rapidly advancing, with a focus on improving the accuracy and reliability of these models. One of the key challenges in this area is the detection and mitigation of hallucinations, which are false or unsupported statements generated by the model. Recent research has made significant progress in addressing this issue, with the development of new methods for uncertainty quantification, hallucination detection, and mitigation. These advances have the potential to improve the trustworthiness and reliability of LLMs, making them more suitable for use in safety-critical applications. Notable papers in this area include 'The Map of Misbelief: Tracing Intrinsic and Extrinsic Hallucinations Through Attention Patterns' and 'CausalGuard: A Smart System for Detecting and Preventing False Information in Large Language Models', which propose innovative approaches to hallucination detection and mitigation.
Advances in Hallucination Detection and Mitigation in Large Language Models
Sources
Bayesian Evaluation of Large Language Model Behavior
Transformers know more than they can tell -- Learning the Collatz sequence
The Map of Misbelief: Tracing Intrinsic and Extrinsic Hallucinations Through Attention Patterns
A Multifaceted Analysis of Negative Bias in Large Language Models through the Lens of Parametric Knowledge
Can LLMs Detect Their Own Hallucinations?
PAS : Prelim Attention Score for Detecting Object Hallucinations in Large Vision--Language Models
CausalGuard: A Smart System for Detecting and Preventing False Information in Large Language Models
Don't Think of the White Bear: Ironic Negation in Transformer Models Under Cognitive Load
Mitigating Length Bias in RLHF through a Causal Lens
Quantifying consistency and accuracy of Latent Dirichlet Allocation
Why is "Chicago" Predictive of Deceptive Reviews? Using LLMs to Discover Language Phenomena from Lexical Cues
What Works for 'Lost-in-the-Middle' in LLMs? A Study on GM-Extract and Mitigations
Collaborative QA using Interacting LLMs. Impact of Network Structure, Node Capability and Distributed Data
SymLoc: Symbolic Localization of Hallucination across HaluEval and TruthfulQA
Failure to Mix: Large language models struggle to answer according to desired probability distributions
Graded strength of comparative illusions is explained by Bayesian inference
COMPASS: Context-Modulated PID Attention Steering System for Hallucination Mitigation
Mathematical Analysis of Hallucination Dynamics in Large Language Models: Uncertainty Quantification, Advanced Decoding, and Principled Mitigation
As If We've Met Before: LLMs Exhibit Certainty in Recognizing Seen Files
SeSE: A Structural Information-Guided Uncertainty Quantification Framework for Hallucination Detection in LLMs
MedBayes-Lite: Bayesian Uncertainty Quantification for Safe Clinical Decision Support