Enhancing Hallucination Detection through Perturbation-Based Synthetic Data Generation in System Responses
This addresses the challenge of expensive and outdated annotation for hallucination detection across domains, though it is incremental as it builds on synthetic data generation methods.
The paper tackled the problem of detecting hallucinations in large language model outputs by generating synthetic data through perturbation-based rewriting of system responses, resulting in a T5-base model that outperformed state-of-the-art zero-shot detectors and existing synthetic methods in accuracy and latency.
Detecting hallucinations in large language model (LLM) outputs is pivotal, yet traditional fine-tuning for this classification task is impeded by the expensive and quickly outdated annotation process, especially across numerous vertical domains and in the face of rapid LLM advancements. In this study, we introduce an approach that automatically generates both faithful and hallucinated outputs by rewriting system responses. Experimental findings demonstrate that a T5-base model, fine-tuned on our generated dataset, surpasses state-of-the-art zero-shot detectors and existing synthetic generation methods in both accuracy and latency, indicating efficacy of our approach.