ConfQA: Answer Only If You Are Confident

2025-06-08

AI Summary

To address hallucination in factual statement generation by large language models (LLMs), this paper proposes ConfQA, a fine-tuning framework centered on a “confidence-triggered answering” mechanism: the model responds only when its confidence exceeds a dynamically adjusted threshold; otherwise, it explicitly states “I am uncertain.” Methodologically, ConfQA integrates three innovations: (1) confidence calibration via suppressive prompting aligned with knowledge graph attributes; (2) a Dual Neural Knowledge framework enabling adaptive synergy between neural representations and symbolic knowledge; and (3) confidence-guided supervised fine-tuning coupled with knowledge distillation. Experiments demonstrate that ConfQA reduces hallucination rates from 20–40% to under 5%, achieves over 95% answer accuracy, exhibits strong cross-domain generalization, and decreases reliance on external retrieval by more than 30%.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

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: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Can we teach Large Language Models (LLMs) to refrain from hallucinating factual statements? In this paper we present a fine-tuning strategy that we call ConfQA, which can reduce hallucination rate from 20-40% to under 5% across multiple factuality benchmarks. The core idea is simple: when the LLM answers a question correctly, it is trained to continue with the answer; otherwise, it is trained to admit"I am unsure". But there are two key factors that make the training highly effective. First, we introduce a dampening prompt"answer only if you are confident"to explicitly guide the behavior, without which hallucination remains high as 15%-25%. Second, we leverage simple factual statements, specifically attribute values from knowledge graphs, to help LLMs calibrate the confidence, resulting in robust generalization across domains and question types. Building on this insight, we propose the Dual Neural Knowledge framework, which seamlessly select between internally parameterized neural knowledge and externally recorded symbolic knowledge based on ConfQA's confidence. The framework enables potential accuracy gains to beyond 95%, while reducing unnecessary external retrievals by over 30%.

Problem

• Reduce LLM hallucination rates below 5%
• Train LLMs to admit uncertainty when unsure
• Balance neural and symbolic knowledge retrieval

Innovation

• Fine-tuning strategy reduces hallucination significantly
• Dampening prompt guides confident answering behavior
• Dual Neural Knowledge framework optimizes accuracy and efficiency