Navigating Uncertainty: Optimizing API Dependency for Hallucination Reduction in Closed-Book Question Answering
Pierre Erbacher Sorbonne Université、Louis Falissar Sorbonne Université \ANDVincent Guigue AgroParisTech、Laure Soulier Sorbonne Université * Authors contributed equally to this work
Abstract
While Large Language Models (LLM) are able to accumulate and restore knowledge, they are still prone to hallucination. Especially when faced with factual questions, LLM cannot only rely on knowledge stored in parameters to guarantee truthful and correct answers. Augmenting these models with the ability to search on external information sources, such as the web, is a promising approach to ground knowledge to retrieve information. However, searching in a large collection of documents introduces additional computational/time costs. An optimal behavior would be to query external resources only when the LLM is not confident about answers. In this paper, we propose a new LLM able to self-estimate if it is able to answer directly or needs to request an external tool. We investigate a supervised approach by introducing a hallucination masking mechanism in which labels are generated using a close book question-answering task. In addition, we propose to leverage parameter-efficient fine-tuning techniques to train our model on a small amount of data. Our model directly provides answers for $78.2\%$ of the known queries and opts to search for $77.2\%$ of the unknown ones. This results in the A
中文速览
大语言模型（LLM）在回答事实性问题时容易产生"幻觉"——给出看似合理却错误的答案，而现有方法要么完全依赖外部检索（每次都调用搜索API，代价高昂），要么完全依赖模型自身的参数知识（易出错）。为此，研究者提出了一种"幻觉遮蔽机制"（Hallucination Masking Mechanism，HalM）：先让模型在闭卷问答任务中自测哪些问题会答错，再把错误答案替换成一个特殊的`<search>`标记作为训练标签，从而让模型学会"有把握就直接回答、没把握就主动调用搜索"。结合参数高效微调方法LoRA，该方案在T5-XXL等模型上实现了对78.2%已知问题直接给出正确答案、对77.2%未知问题主动触发搜索，整体仅需调用外部API约62%的时间，而非几乎每次都调用。这项工作的意义在于，它让LLM真正学会了"知之为知之，不知则求助"，在保证回答质量的同时大幅降低了对外部资源的依赖，为构建更高效、更可靠的知识问答系统提供了可行路径。
原文 arXiv:2401.01780;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2401.01780v1