Recent breakthroughs in Large Language Models (LLMs) have revealed remarkable generative capabilities and emerging self-regulatory mechanisms, including self-correction and self-rewarding. However, current detoxification techniques rarely exploit these built-in abilities; instead, they rely on external modules, labor-intensive data annotation, or human intervention –factors that hinder scalability and consistency. In this paper, we introduce a fully self-reflective detoxification framework that harnesses the inherent capacities of LLMs to detect, correct toxic content, and refine LLMs without external modules and data annotation. Specifically, we propose a Toxic Signal Detector—an internal self-identification mechanism, coupled with a systematic intervention process to transform toxic text into its non-toxic counterpart. This iterative procedure yields a contrastive detoxification dataset used to fine-tune the model, enhancing its ability for safe and coherent text generation. Experiments on benchmark datasets such as DetoxLLM and ParaDetox show that our method achieves better detoxification performance than state-of-the-art methods while preserving semantic fidelity. By obviating the need for human intervention or external components, this paper reveals the intrinsic self-detoxification ability of LLMs, offering a consistent and effective approach for mitigating harmful content generation. Ultimately, our findings underscore the potential for self-regulated language models and more responsible text generation systems. Code: https://github.com/KaituoZhang/SRD Warning: this paper may contain offensive content.

Large language models (LLMs) remain vulnerable to jailbreak attacks that exhibit diverse prompts and rapidly evolving strategies. Existing defenses often operate at the level of prompts or predefined attack methods, limiting their ability to generalize to unseen attacks that share similar underlying structure but differ in wording. In this paper, we investigate whether seemingly different jailbreak attacks exhibit shared SAE feature co-activation patterns, and propose a motif-based framework that models attacks through these recurring patterns. Specifically, we first extract sparse autoencoder (SAE) features from jailbreak queries and discover motifs as recurring co-activation patterns via nonnegative matrix factorization. We show that motifs capture stable relationships across attack methods and generalize to unseen attacks. Building on this structure, we synthesize new jailbreak queries by composing motif-guided attack patterns. We then use the generated queries to construct preference-training data for Direct Preference Optimization (DPO). Experiment results show that SAE motifs serve as a stable and transferable interface for understanding and detecting jailbreak attacks, as well as enabling effective defenses that generalize beyond fixed prompt templates. Training target models with motif-guided synthetic data improves robustness to unseen attacks with modest changes in utility. Our work provides a practical methodology for advancing data-centric LLM safety through structured modeling of jailbreak attacks.

Although weight quantization helps large language models (LLMs) in resource-constrained environments, its influence on the uncertainty calibration remains unexplored. To bridge this gap, we presents a comprehensive investigation of uncertainty calibration for quantized LLMs in this work. Specifically, we propose an analytic method to estimate the upper bound of calibration error (UBCE) for LLMs. Our method separately discusses the calibration error of the model’s correct and incorrect predictions, indicating a theoretical improvement of calibration error caused by the weight quantization. Our study demonstrates that quantized models consistently exhibit worse calibration performance than full-precision models, supported by consistent analysis across multiple LLMs and datasets. To address the calibration issues of quantized models, we propose a novel method of post calibration for recovering the calibration performance of quantized models through soft-prompt tuning. Specifically, we inject soft tokens to quantized models after the embedding layers, and optimize these tokens to recover the calibration error caused by the weight quantization. Experimental results on multiple datasets demonstrate our effectiveness in improving the uncertainty calibration of quantized LLMs, facilitating more reliable weight quantization in resource-constrained environments.

Large Language Models (LLMs) are becoming essential tools for various natural language processing tasks but often suffer from generating outdated or incorrect information. Retrieval-Augmented Generation (RAG) addresses this issue by incorporating external, real-time information retrieval to ground LLM responses. However, the existing RAG systems frequently struggle with the quality of retrieval documents, as irrelevant or noisy documents degrade performance, increase computational overhead, and undermine response reliability. To tackle this problem, we propose Multi-Agent Filtering Retrieval-Augmented Generation (MAIN-RAG), a training-free RAG framework that leverages multiple LLM agents to collaboratively filter and score retrieved documents. Specifically, MAIN-RAG introduces an adaptive filtering mechanism that dynamically adjusts the relevance filtering threshold based on score distributions, effectively minimizing noise while maintaining high recall of relevant documents. The proposed approach leverages inter-agent consensus to ensure robust document selection without requiring additional training data or fine-tuning. Experimental results across four QA benchmarks demonstrate that MAIN-RAG consistently outperforms traditional RAG approaches, achieving a 2–11% improvement in answer accuracy while reducing the number of irrelevant retrieved documents. Quantitative analysis further reveals that our approach achieves superior response consistency and answer accuracy over baseline methods, offering a competitive and practical alternative to training-based solutions.