DSCC-HS: A Dynamic Self-Reinforcing Framework for Hallucination Suppression in Large Language Models

2025-09-17

AI Summary

LLM hallucinations severely hinder their trustworthy deployment. This paper proposes a dynamic intervention framework that enhances factual consistency during autoregressive decoding without modifying the target model. Methodologically, it introduces (1) a dual-agent collaboration mechanism—comprising a Factuality-Aware Planner (FAP) and a Hallucination-Detection Processor (HDP)—integrating retrieval-augmented generation with dual-process cognitive theory; (2) a dynamic self-amplifying calibration module, employing a lightweight adversarially trained agent to generate logits-level steering vectors in real time, enabling plug-and-play integration; and (3) a logits-differential control technique for fine-grained factual constraint enforcement. Evaluated on TruthfulQA, the framework achieves 99.2% factual accuracy; on BioGEN, it attains a state-of-the-art FActScore of 46.50—substantially outperforming existing approaches.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Adversarial Agents

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
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Large Language Model (LLM) hallucination is a significant barrier to their reliable deployment. Current methods like Retrieval-Augmented Generation (RAG) are often reactive. We introduce **Dynamic Self-reinforcing Calibration for Hallucination Suppression (DSCC-HS)**, a novel, proactive framework that intervenes during autoregressive decoding. Inspired by dual-process cognitive theory, DSCC-HS uses a compact proxy model, trained in adversarial roles as a Factual Alignment Proxy (FAP) and a Hallucination Detection Proxy (HDP). During inference, these proxies dynamically steer a large target model by injecting a real-time steering vector, which is the difference between FAP and HDP logits, at each decoding step. This plug-and-play approach requires no modification to the target model. Our experiments on TruthfulQA and BioGEN show DSCC-HS achieves state-of-the-art performance. On TruthfulQA, it reached a 99.2% Factual Consistency Rate (FCR). On the long-form BioGEN benchmark, it attained the highest FActScore of 46.50. These results validate DSCC-HS as a principled and efficient solution for enhancing LLM factuality.

Problem

• Proactively suppressing hallucinations in large language models
• Dynamic intervention during autoregressive decoding process
• Enhancing factual consistency without modifying target model

Innovation

• Dynamic self-reinforcing framework for hallucination suppression
• Uses dual proxy models for real-time steering
• Plug-and-play approach without model modification