Safety alignment in Large Language Models (LLMs) remains highly fragile during fine-tuning, where even benign adaptation can degrade pre-trained refusal behaviors and enable harmful responses. Existing defenses typically constrain either weights or activations in isolation, without considering their coupled effects on safety. In this paper, we first theoretically demonstrate that constraining either weights or activations alone is insufficient for safety preservation. To robustly preserve safety alignment, we propose Coupled Weight and Activation Constraints (CWAC), a novel approach that simultaneously enforces a precomputed safety subspace on weight updates and applies targeted regularization to safety-critical features identified by sparse autoencoders. Extensive experiments across four widely used LLMs and diverse downstream tasks show that CWAC consistently achieves the lowest harmful scores with minimal impact on fine-tuning accuracy, substantially outperforming strong baselines even under high harmful data ratios.

Temporal knowledge graph forecasting(TKGF) asks a model to rank the mostplausible future entity for a query such as(s, r, ?, t) from historical events. Recenttraining-free methods use large languagemodels (LLMs) for this task, but their accuracydepends heavily on which past events areshown in the prompt under a tight contextbudget. We present LANTERN, a training-freeprompting framework that addresses thisbottleneck by combining two complementaryviews of history: a long-window strengthscore for stable interaction patterns anda short-window novelty score for suddenchanges. LANTERN first filters unhelpfulevents, then selects a compact evidence setwith Pareto-greedy selection, and finally addsone structure-aware analogical demonstration.Across ICEWS14, ICEWS05-15, ICEWS18,and GDELT, LANTERN consistently outperforms the state-of-the-art training-free baselineAnRe under the same backbone and 2-hopcandidate protocol, improving Hits@1 by upto 2.5 points and MRR by up to 1.2 points.

Embedding-as-a-Service (EaaS) has emerged as a critical paradigm for commercializing large language models (LLMs). However, existing backdoor watermarking techniques are fundamentally limited to "zero-bit" detection, which prevents user-level traceability in multi-user EaaS scenarios. To address these limitations, we propose RShield, a multi-bit backdoor watermarking that enables reliable user-level attribution of LLMs for EaaS under model extraction attacks. RShield integrates Reed-Solomon error-correcting codes with orthogonal feature mapping to introduce highly-structured redundancy, constructing fault-tolerant symbol sequences for multi-bit watermark space, thereby staying recoverable even after aggressive extraction noise condition.To mitigate semantic distortion under the interference of noise channel, RShield employs a lightweight Adapter to adaptively inject multi-bit watermarks in the feature space, preserving the quality of EaaS while achieving a user-level traceability.Extensive experiments on four NLP benchmarks demonstrate that RShield efficiently achieves 100% multi-bit watermark recovery and high semantic fidelity under model extraction attacks compared to existing methods, while significantly reducing the degradation of watermarking on downstream task performance.

LLM-based Multi-agent systems (MAS) have shown strong capabilities across a wide range of domains. Their success largely hinges on the collaboration topology design, which has emerged as a central research focus in the automated MAS design.However, existing approaches are fundamentally constrained by their reliance on homogeneous LLMs, which significantly limits overall system intelligence.In response to this limitation, we for the first time propose the concept of **Automated Design of Heterogeneous-LLMs-based MAS (ADHM)**.ADHM sheds light on a promising avenue for advancing collective intelligence, which focuses on the automated design of cost-effective MAS composed of diverse LLMsand roles to suit various queries.Toward this challenging goal, we propose **Hetero-Designer**, a novel pipeline that efficiently encodes intricate dependencies among queries, LLMs and roles through a novel Binary-Star Transformer and constructs Hetero-MAS in an autoregressive graph generation process. Extensive experiments demonstrate that **Hetero-Designer** is: (1) high-performing on various benchmarks, (2) economical in reducing overhead, (3) extensible to unseen LLMs and roles.

In this paper, we focus on the challenging yet practical problem of Continual Few-shot Relation Extraction (CFRE), which involves extracting relations in the continuous and iterative arrival of new data with only a few labeled examples. The main challenges in CFRE are overfitting due to few-shot learning and catastrophic forgetting caused by continual learning. To address these problems, we propose a novel framework called RK2DA, which seamlessly integrates prototype-based data augmentation and relational knowledge distillation. Specifically, RK2DA generates pseudo data by introducing Gaussian noise to the prototype embeddings and utilizes a novel two-phase multi-teacher relational knowledge distillation method to transfer various knowledge from different embedding spaces. Experimental results on the FewRel and TACRED datasets demonstrate that our method outperforms the state-of-the-art baselines.

Large Language Models (LLMs) have shown impressive capabilities but still suffer from the issue of hallucinations. A significant type of this issue is the false premise hallucination, which we define as the phenomenon when LLMs generate hallucinated text when confronted with false premise questions. In this paper, we perform a comprehensive analysis of the false premise hallucination and elucidate its internal working mechanism: a small subset of attention heads (which we designate as false premise heads) disturb the knowledge extraction process, leading to the occurrence of false premise hallucination. Based on our analysis, we propose FAITH (False premise Attention head constraIining for miTigating Hallucinations), a novel and effective method to mitigate false premise hallucinations. It constrains the false premise attention heads during the model inference process. Impressively, extensive experiments demonstrate that constraining only approximately 1% of the attention heads in the model yields a notable increase of nearly 20% of model performance.

Adjusting the outdated behaviors of large langugae models (LLMs) after deployment remains a significant challenge. It motivates the model editing research, which is however mainly explored in a restricted task form with triple-based edit requests. Recent works have initiated a transition to a more practical and unified editing task that takes free-form text as edit requests. However, there are gaps in nuanced benchmark designs and re-evaluation of existing methods. To bridge the gaps, we introduce a multi-level benchmark for free text model editing (MULFE). The benchmark categorizes probe queries into three levels of generalization, ranging from basic literal memory to deeper understanding and reasoning. Based on the benchmark, we conduct extensive experiments across various base models, edit sizes, and editing methods, including adaptations of mainstream locate-and-edit and hypernetwork methods. The results highlight the inconsistent behaviors of edited models on different generalization levels. Higher-level generalization remains a significant challenge. Based on the findings, we propose SIDE, a simple yet effective method based on in-context distillation to enhance the generalization performance. The benchmark dataset and evaluation scripts are publicly available at http://github.com/wchrepo/mulfe.

Large language models exhibit high-level commonsense reasoning abilities, especially with enhancement methods like Chain-of-Thought (CoT). However, we find these CoT-like methods lead to a considerable number of originally correct answers turning wrong, which we define as the Toxic CoT problem. To interpret and mitigate this problem, we first utilize attribution tracing and causal tracing methods to probe the internal working mechanism of the LLM during CoT reasoning. Through comparisons, we prove that the model exhibits information loss from the question over the shallow attention layers when generating rationales or answers. Based on the probing findings, we design a novel method called RIDERS (Residual decodIng and sERial-position Swap), which compensates for the information deficit in the model from both decoding and serial-position perspectives. Through extensive experiments on multiple commonsense reasoning benchmarks, we validate that this method not only significantly eliminates Toxic CoT problems (decreased by 23.6%), but also effectively improves the model’s overall commonsense reasoning performance (increased by 5.5%).

Joint entity and relation extraction aims to extract relation triplets from plain text directly. Prior work leverages Sequence-to-Sequence (Seq2Seq) models for triplet sequence generation. However, Seq2Seq enforces an unnecessary order on the unordered triplets and involves a large decoding length associated with error accumulation. These methods introduce exposure bias, which may cause the models overfit to the frequent label combination, thus limiting the generalization ability. We propose a novel Sequence-to-Unordered-Multi-Tree (Seq2UMTree) model to minimize the effects of exposure bias by limiting the decoding length to three within a triplet and removing the order among triplets. We evaluate our model on two datasets, DuIE and NYT, and systematically study how exposure bias alters the performance of Seq2Seq models. Experiments show that the state-of-the-art Seq2Seq model overfits to both datasets while Seq2UMTree shows significantly better generalization. Our code is available at https://github.com/WindChimeRan/OpenJERE.

The multiple relation extraction task tries to extract all relational facts from a sentence. Existing works didn’t consider the extraction order of relational facts in a sentence. In this paper we argue that the extraction order is important in this task. To take the extraction order into consideration, we apply the reinforcement learning into a sequence-to-sequence model. The proposed model could generate relational facts freely. Widely conducted experiments on two public datasets demonstrate the efficacy of the proposed method.

The relational facts in sentences are often complicated. Different relational triplets may have overlaps in a sentence. We divided the sentences into three types according to triplet overlap degree, including Normal, EntityPairOverlap and SingleEntiyOverlap. Existing methods mainly focus on Normal class and fail to extract relational triplets precisely. In this paper, we propose an end-to-end model based on sequence-to-sequence learning with copy mechanism, which can jointly extract relational facts from sentences of any of these classes. We adopt two different strategies in decoding process: employing only one united decoder or applying multiple separated decoders. We test our models in two public datasets and our model outperform the baseline method significantly.