Smoothing Out Hallucinations: Mitigating LLM Hallucination with Smoothed Knowledge Distillation

2025-02-16

AI Summary

To address the pervasive hallucination problem—i.e., factual inconsistencies—in large language model (LLM) generations, this paper proposes a knowledge distillation method based on smoothed soft labels. During supervised fine-tuning, the approach leverages uncertainty-aware soft targets produced by a teacher model to mitigate overconfidence and factual detachment in the student model. This work is the first to systematically integrate soft label smoothing into hallucination mitigation for LLMs, explicitly modeling uncertainty in generative language modeling while preserving general-purpose capabilities. Experiments across diverse architectures—including Llama, Qwen, and Phi—demonstrate consistent improvements: average hallucination rates decrease by 23.6% on authoritative benchmarks such as SummEval and QAGS, with no degradation in performance on general NLP tasks, confirming strong cross-architecture generalizability.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Diffusion Models for Vision

Application Category

• 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
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large language models (LLMs) often suffer from hallucination, generating factually incorrect or ungrounded content, which limits their reliability in high-stakes applications. A key factor contributing to hallucination is the use of hard labels during training, which enforce deterministic supervision, encourage overconfidence, and disregard the uncertainty inherent in natural language. To address this, we propose mitigating hallucination through knowledge distillation (KD), where a teacher model provides smoothed soft labels to a student model, reducing overconfidence and improving factual grounding. We apply KD during supervised finetuning on instructional data, evaluating its effectiveness across LLMs from different families. Experimental results on summarization benchmarks demonstrate that KD reduces hallucination compared to standard finetuning while preserving performance on general NLP tasks. These findings highlight KD as a promising approach for mitigating hallucination in LLMs and improving model reliability.

Problem

• Mitigate LLM hallucination via knowledge distillation.
• Use smoothed soft labels to reduce overconfidence.
• Improve factual grounding in large language models.

Innovation

• Knowledge distillation mitigates hallucinations
• Soft labels reduce model overconfidence
• Supervised finetuning with instructional data