HICD: Hallucination-Inducing via Attention Dispersion for Contrastive Decoding to Mitigate Hallucinations in Large Language Models
This addresses the issue of inaccurate or incorrect outputs in LLMs for users relying on reliable text generation, though it is incremental as it builds on existing contrastive decoding methods.
The paper tackled the problem of hallucinations in large language models by introducing HICD, a method that induces hallucinations via attention dispersion for contrastive decoding, resulting in significant improvements in contextual faithfulness and factuality across tasks like context completion and question answering.
Large Language Models (LLMs) often generate hallucinations, producing outputs that are contextually inaccurate or factually incorrect. We introduce HICD, a novel method designed to induce hallucinations for contrastive decoding to mitigate hallucinations. Unlike existing contrastive decoding methods, HICD selects attention heads crucial to the model's prediction as inducing heads, then induces hallucinations by dispersing attention of these inducing heads and compares the hallucinated outputs with the original outputs to obtain the final result. Our approach significantly improves performance on tasks requiring contextual faithfulness, such as context completion, reading comprehension, and question answering. It also improves factuality in tasks requiring accurate knowledge recall. We demonstrate that our inducing heads selection and attention dispersion method leads to more "contrast-effective" hallucinations for contrastive decoding, outperforming other hallucination-inducing methods. Our findings provide a promising strategy for reducing hallucinations by inducing hallucinations in a controlled manner, enhancing the performance of LLMs in a wide range of tasks.