CLJul 10, 2024

Knowledge Overshadowing Causes Amalgamated Hallucination in Large Language Models

arXiv:2407.08039v131 citationsh-index: 24
AI Analysis

This addresses a critical issue for users relying on LLMs for knowledge-intensive tasks by revealing a systematic cause of hallucination, though it is incremental as it builds on existing understanding of data imbalance effects.

The paper tackles the problem of hallucination in large language models, specifically identifying 'knowledge overshadowing' where multiple query conditions cause amalgamated false outputs, and shows that hallucination rates increase with data imbalance and condition length, achieving up to 82% F1 for anticipation and 11.2% to 39.4% control with proposed methods.

Hallucination is often regarded as a major impediment for using large language models (LLMs), especially for knowledge-intensive tasks. Even when the training corpus consists solely of true statements, language models still generate hallucinations in the form of amalgamations of multiple facts. We coin this phenomenon as ``knowledge overshadowing'': when we query knowledge from a language model with multiple conditions, some conditions overshadow others, leading to hallucinated outputs. This phenomenon partially stems from training data imbalance, which we verify on both pretrained models and fine-tuned models, over a wide range of LM model families and sizes.From a theoretical point of view, knowledge overshadowing can be interpreted as over-generalization of the dominant conditions (patterns). We show that the hallucination rate grows with both the imbalance ratio (between the popular and unpopular condition) and the length of dominant condition description, consistent with our derived generalization bound. Finally, we propose to utilize overshadowing conditions as a signal to catch hallucination before it is produced, along with a training-free self-contrastive decoding method to alleviate hallucination during inference. Our proposed approach showcases up to 82% F1 for hallucination anticipation and 11.2% to 39.4% hallucination control, with different models and datasets.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes