Aili Chen, Chengyu Du, Jiangjie Chen, Jinghan Xu, Yikai Zhang, Siyu Yuan, Zulong Chen, Liangyue Li, Yanghua Xiao. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Large Language Models (LLMs) achieve remarkable success, but their static nature leads to inherent limitations and persistent mistakes in dynamic real-world scenarios. While Continual Instruction Tuning (CIT) and Continual Pre-training (CPT) are primary continual learning approaches, they struggle with scalable knowledge acquisition and maintaining model capabilities. To address these, we propose the Continue Evolving from Mistakes (CEM) method, a novel and data-efficient framework for continuous LLM evolution. Inspired by human learning, CEM establishes an iterative process: it efficiently collects targeted CPT data by robustly identifying LLM mistakes and uncertainties (via an Ambiguity-Aware Knowledge Collection (AAKC) algorithm), and employs a novel joint training paradigm that leverages CIT and CPT to assimilate knowledge efficiently while maintaining existing capabilities and mitigating catastrophic forgetting. Extensive experiments confirm CEM's effectiveness, yielding substantial accuracy gains for multiple models, increasing accuracy by up to 29.63%. Code and datasets are available on GitHub https://anonymous.4open.science/r/cem-BB25.

Yizhou Ying, Geng Zhang, Cui Danxin, Chengyu Du, Guanglei Yue, Sihang Jiang, Jiaqing Liang, Yifei Fu, Hailin Hu, Yanghua Xiao. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Recent advancements in large language models (LLMs) have demonstrated that progressive refinement, rather than providing a single answer, results in more accurate and thoughtful outputs. However, existing methods often rely heavily on supervision signals to evaluate previous responses, making it difficult to effectively assess output quality in more open-ended scenarios. Additionally, these methods are typically designed for specific tasks, which limits their generalization to new domains. To address these limitations, we propose Progressive Thought Refinement (PTR), a framework that enables LLMs to progressively refine their responses. PTR operates in two phases: (1) Thought data construction stage: We propose a weak and strong model collaborative selection strategy to build a high-quality progressive refinement dataset to ensure logical consistency from thought to answers, and the answers are gradually refined in each round. (2) Thought-Mask Fine-Tuning Phase: We design a training structure to mask the "thought" and adjust loss weights to encourage LLMs to refine prior thought, teaching them to implicitly understand "how to improve" rather than "what is correct." Experimental results show that PTR significantly enhances LLM performance across ten diverse tasks (avg. from 49.6% to 53.5%) without task-specific fine-tuning. Notably, in more open-ended tasks, LLMs also demonstrate substantial improvements in the quality of responses beyond mere accuracy, suggesting that PTR truly teaches LLMs to self-improve over time. Our work is now open-source. https://github.com/cydu24/Progressive-Thought-Refinement

The password is a sensitive personal information which renders centralized plaintext training ethically and legally untenable under prevailing regulations. Existing privacy-preserving techniques (e.g., federated learning, homomorphic encryption) are characterised by computational overhead, performance degradation, and inadequate task-specific optimisation for password data. To address these limitations, a privacy-preserving multi-task password guessing framework, termed P³G, is proposed. The model under consideration adopts a shared-private architecture with three tailored innovations. Firstly, a Structure-Aware Differential Privacy (SADP) mechanism is designed which injects Gaussian noise modulated by password character type and network depth to preserve semantic consistency. Furthermore, a Complexity-Aware Curriculum Learning (CACL) mechanism is proposed to be adopted as a guiding principle for training, with the objective of transitioning from simple to complex passwords via entropy-based temperature scheduling. Finally, Strength-Aware Task Weighting (SATW) mechanism is a dynamic balancing of tasks using password strength metrics. Experiments on real-world datasets (CSDN, RockYou, 000webhost) demonstrate that P³G reduces perplexity by 7.36 percent for generative password guessing and improves masked recovery accuracy by 3.07 percent. Furthermore, it achieves a 26.06 percent lower task conflict (TC) and a privacy leakage index (PLI) of only 0.093. To the best of our knowledge, this is the first work to tailor privacy mechanisms explicitly for multi-task password modeling, offering a new perspective for password research.