Knowledge Editing (KE) aims to update specific facts in large language models (LLMs) without full retraining. Prior efforts sought to tune the knowledge layers of LLMs, achieving near-perfect performance in teacher-forced evaluations. However, under realistic autoregressive generation evaluation, their effectiveness remains limited by two core challenges: (1) even when methods successfully inject new knowledge, pre-trained capabilities can degrade, and (2) more critically, the injected parametric knowledge is not reliably expressed at inference time. To this end, we formalize knowledge editing as a two-stage problem and propose EtCon, an edit-then-consolidate paradigm that couples targeted edits with post-edit consolidation. Within EtCon, the edit stage applies Targeted Proximal Supervised Fine-Tuning (TPSFT) to update parametric knowledge with minimal disruption to pre-trained capabilities. The consolidation stage applies Group Relative Policy Optimization (GRPO) to align the model’s autoregressive generation with the edited parametric knowledge. Extensive experiments demonstrate that our instantiation (TPSFT + GRPO) improves editing reliability and real-world generalization while preserving pre-trained capabilities, and that EtCon consistently improves existing editing methods when instantiated with them.

While large language models (LLMs) show immense potential in specialized domains, their adaptation is often hindered by limited access to domain-specific data and the high computational cost of model training. To address these challenges, we propose a lightweight knowledge injection framework. Instead of sharing raw data directly, our approach fine-tunes a small model and transfers its knowledge vectors to the target LLM for knowledge injection, thereby reducing data exposure and computational overhead. To mitigate the knowledge loss incurred during the injection process, we further introduce an iterative knowledge calibration strategy. We theoretically bound the error of the transformation matrix and prove the stability and linear convergence of the calibration process. Comprehensive experiments demonstrate that our method significantly outperforms parameter-efficient baselines and achieves performance comparable to full-parameter fine-tuning, while consuming only 28% of the GPU memory and 3% of the raw training data. Code is available at https://github.com/Yuhanleeee/LiSKI.

Despite the remarkable success of Large Language Models (LLMs) in Machine Translation (MT), the scarcity of high-quality parallel corpora and the prohibitive cost of their acquisition constrain scalability. To this end, we propose Learning to Translate by Translating (LTT), an LLM-driven dual-learning framework that enables autonomous translation, achieving an 80.42% performance improvement over the base model. By adapting the cycle-consistency principle to the generative paradigm, LTT eliminates the need for parallel data. It employs a robust semantic-aware reward function that balances adequacy with reconstruction fidelity, effectively mitigating the reward hacking issues inherent in traditional unsupervised MT. Relying solely on monolingual data, our 8B model consistently outperforms significantly larger models (70B+) in low-resource settings and achieves parity with state-of-the-art supervised baselines on mainstream benchmarks. LTT thus offers a scalable, data-efficient paradigm for autonomous machine translation.

Vision-Language Models (VLMs) based GUI agents stand to benefit significantly from online reinforcement learning (RL). However, their training is bottlenecked by two fundamental issues: current data synthesis methods for GUI Agents rely on specific environments and struggle to generate diverse data, while existing evaluators either suffer from limited scalability or provide inaccurate and unreliable reward signals. To overcome these challenges, we introduce GSAR (Goal-State-Anchor Reward), a RL reward framework that supports scalable task generation and delivers reliable reward signals for stable and efficient policy optimization. Our approach features self-evolving data synthesis, which produces multiple environments through task execution and generates diverse tasks and goal states. Complementing this, a state-anchor mechanism automatically annotates task-relevant UI elements in successful goal states as reference anchors. During RL training, these reference anchors provide accurate, scalable reward signals that substantially enhance efficiency. Extensive evaluations demonstrate that our framework achieves over 90% accuracy on offline trajectory verification and performs closest to rule-based methods. Furthermore, agents trained using our reward framework exhibit strong performance on both AndroidWorld and our constructed benchmark, establishing a scalable approach for GUI agent training.

Recent advancements in large language models (LLMs) have given rise to the emergence of role-playing agents (RPAs). The development of high-quality dialogue datasets is critical for advancing RPAs. However, existing datasets have two main issues: (1) the bias between query distributions and real-world user language usage, and (2) the challenge of ensuring responses accurately reflect character traits.To address these issues, we propose CharacterCraft, a novel framework designed for practical RPAs, comprising a tailored Chinese role-playing dataset and a robust evaluation method. First, we develop a specialized model for Chinese dialogue extraction, achieving state-of-the-art performance. Using this model, we then extract a large amount of character dialogue from novels, ensuring high data quality (issue 2).To mitigate the literature-reality dialogue bias in extracted dialogue (issue 1), we introduce an iterative augmentation-reconstruction method, which revises queries to better align with common language usage. Additionally, we propose a context-aware memory retrieval module for fine-grained alignment with the character and introduce a reference-guided LLM-as-a-judge evaluation method for more reliable assessments by comparing their responses to source material dialogues.Our automated pipeline produces a large-scale high-quality Chinese role-playing dataset with 21,392 samples and 121,418 utterances. The experimental results demonstrate the effectiveness of our framework and reveal the limitations of existing RPAs when faced with diverse scenes.Our repository is at https://github.com/yin214/CharacterCraft.

Multi-style outline controllable generation is crucial for multiple applications, including document semantic structuring and retrieval-augmented generation.The great success of preference alignment approaches encourages their application in controllable generation tasks.However, these attempts encounter several limitations: (1) response pair requirements, (2) substantial computation costs, and (3) insufficient exploitation of fine-grained preference signals.To address these problems, we propose a token-level preference self-alignment optimization, named TKPO, for outline controllable generation. TKPO extends the Bradley-Terry model from pair-wise to list-wise comparison, which is further applied at the token level for fine-grained preference signal utilization. In comparison to the representative methods, e.g., DPO, TKPO does not require response pairs; instead, we propose a controllable attributes-driven method to construct reject samples for self-alignment. Additionally, TKPO optimizes only the base model, thereby avoiding additional memory usage and substantial computational costs.We curate two outline controllable generation datasets with regard to language style and level-of-detail.Extensive experiments demonstrate that TKPO outperforms DPO by up to 19.28% in performance while requiring only 56.25% in training time.We release the code and datasets resources at https://github.com/WHUIR/TKPO.

Recent advancements in AI-generated content (AIGC) have heightened concerns about harmful outputs, such as misinformation and malicious misuse.Existing detection methods face two key limitations:(1) lacking real-world AIGC scenarios and corresponding risk datasets, and(2) both traditional and multimodal large language models (MLLMs) struggle to detect risks in AIGC.Towards this end, we introduce AIGuard, the first benchmark for AIGC risk detection in real-world e-commerce. It includes 253,420 image-text pairs (i.e., the risk content and risk description) across four critical categories: abnormal body, violating physical laws, misleading or illogical context, and harmful or problematic message.To effectively detect these risks, we propose distilling text annotations into dense soft prompts and identifying risk content through image soft prompt matching during inference.Experiments on the benchmark show that this method achieves a 9.68% higher recall than leading multimodal models while using only 25% of the training resources and improving inference speed by 37.8 times.For further research, our benchmark and code are available at https://github.com/wenh-zhang/aiguard-dataset.

Honest alignment refers to the ability of a language model to truthfully convey its knowledge limitations by appropriately refusing to answer questions when it lacks sufficient information. Existing solutions, such as prompt engineering and fine-tuning, face limitations: the former provides only marginal improvements, while the latter struggles to enhance honesty when annotated data is scarce.To overcome the above limitations, we propose , a novel framework that enhances honesty through weak-to-strong generalization. Specifically, we train the strong LLMs under weak model supervision to improve their honesty. For the weak model, we employ a learning-to-rank strategy to train a “honest head”, which learns to select the most honest response among model’s outputs generated through beam search. For the strong LLM, we leverage the self-labeled dataset to update its parameters. Our proposal requires only minimal training data to train the weak honest model, yet achieve decent performance for labeling data. In addition, it enables the strong LLMs to have the capabilities to generalize even facing with the flawed label data. Extensive experiments show significantly boosts honest alignment in large models even with limited labeled data. Our code is available at https://github.com/zewanfaan/WHAT_Honesty.

We solve the challenging document-level event extraction problem by proposing a joint exaction methodology that can avoid inefficiency and error propagation issues in classic pipeline methods. Essentially, we address the three crucial limitations in existing studies. First, the autoregressive strategy of path expansion heavily relies on the orders of argument role. Second, the number of events in documents must be specified in advance. Last, unexpected errors usually exist when decoding events based on the entity-entity adjacency matrix. To address these issues, this paper designs a Token-Token Bidirectional Event Completed Graph (TT-BECG) in which the relation eType-Role1-Role2 serves as the edge type, precisely revealing which tokens play argument roles in an event of a specific event type. Exploiting the token-token adjacency matrix of the TT-BECG, we develop an edge-enhanced joint document-level event extraction model. Guided by the target token-token adjacency matrix, the predicted token-token adjacency matrix can be obtained during the model training. Then, extracted events and event records in a document are decoded based on the predicted matrix, including the graph structure and edge type decoding. Extensive experiments are conducted on two public datasets, and the results confirm the effectiveness of our method and its superiority over the state-of-the-art baselines.