Improving Factuality in Large Language Models via Decoding-Time Hallucinatory and Truthful Comparators

2024-08-22

arXiv.org

AI Summary

Large language models (LLMs) frequently generate factually inconsistent hallucinations, undermining their reliability. To address this, we propose a decoding-time dual-comparator framework—termed the Hallucination/Truth Comparator—that dynamically constrains the token prediction distribution during autoregressive generation, enabling factuality calibration without fine-tuning or architectural modification. Our approach innovatively integrates an instruction-prototype-guided Mixture-of-Experts (MoE) mechanism with a logit-difference comparison strategy, establishing a task-agnostic factual constraint paradigm. Extensive experiments across diverse downstream tasks demonstrate substantial improvements in response factuality and overall performance, while preserving the original model’s capabilities. The method is fully plug-and-play, requiring no retraining or parameter updates.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Despite their remarkable capabilities, Large Language Models (LLMs) are prone to generate responses that contradict verifiable facts, i.e., unfaithful hallucination content. Existing efforts generally focus on optimizing model parameters or editing semantic representations, which compromise the internal factual knowledge of target LLMs. In addition, hallucinations typically exhibit multifaceted patterns in downstream tasks, limiting the model's holistic performance across tasks. In this paper, we propose a Comparator-driven Decoding-Time (CDT) framework to alleviate the response hallucination. Firstly, we construct hallucinatory and truthful comparators with multi-task fine-tuning samples. In this case, we present an instruction prototype-guided mixture of experts strategy to enhance the ability of the corresponding comparators to capture different hallucination or truthfulness patterns in distinct task instructions. CDT constrains next-token predictions to factuality-robust distributions by contrasting the logit differences between the target LLMs and these comparators. Systematic experiments on multiple downstream tasks show that our framework can significantly improve the model performance and response factuality.

Problem

• Large Language Models
• Factuality
• Reliability

Innovation

• Comparator-Driven Timing (CDT)
• Statement Discrimination
• Enhanced Accuracy and Fidelity