Understanding events necessitates grasping their temporal context, which is often not explicitly stated in natural language. For example, it is not a trivial task for a machine to infer that a museum tour may last for a few hours, but can not take months. Recent studies indicate that even advanced large language models (LLMs) struggle in generating text that require reasoning with temporal commonsense due to its infrequent explicit mention in text. Therefore, automatically mining temporal commonsense for events enables the creation of robust language models. In this work, we investigate the capacity of LLMs to extract temporal commonsense from text and evaluate multiple experimental setups to assess their effectiveness. Here, we propose a temporal commonsense extraction pipeline that leverages LLMs to automatically mine temporal commonsense and use it to construct TComQA, a dataset derived from SAMSum and RealNews corpora. TComQA has been validated through crowdsourcing and achieves over 80\% precision in extracting temporal commonsense. The model trained with TComQA also outperforms an LLM fine-tuned on existing dataset of temporal question answering task.

Recommendation systems have been essential for both user experience and platform efficiency by alleviating information overload and supporting decision-making. Traditional methods, i.e., content-based filtering, collaborative filtering, and deep learning, have achieved impressive results in recommendation systems. However, the cold-start and sparse-data scenarios are still challenging to deal with. Existing solutions either generate pseudo-interaction sequence, which often introduces redundant or noisy signals, or rely heavily on semantic similarity, overlooking dynamic shifts in user motivation. To address these limitations, this paper proposes a novel recommendation framework, termed M-$LLM^3$REC, which leverages large language models for deep motivational signal extraction from limited user interactions. M-$LLM^3$REC comprises three integrated modules: the Motivation-Oriented Profile Extractor (MOPE), Motivation-Oriented Trait Encoder (MOTE), and Motivational Alignment Recommender (MAR). By emphasizing motivation-driven semantic modeling, M-$LLM^3$REC demonstrates robust, personalized, and generalizable recommendations, particularly boosting performance in cold-start situations in comparison with the state-of-the-art frameworks.

Large Language Models (LLMs) have shown great potential in reasoning tasks through test-time scaling methods like self-consistency with majority voting. However, this approach often leads to diminishing returns in accuracy and high computational overhead. To address these challenges, we introduce Deep Think with Confidence (DeepConf), a simple yet powerful method that enhances both reasoning efficiency and performance at test time. It requires no additional model training or hyperparam-eter tuning and can be seamlessly integrated into existing serving frameworks. We evaluate DeepConf across a variety of reasoning tasks and the latest open-source models, including Qwen 3 and GPT -OSS series. Notably, on challenging benchmarks such as AIME 2025, DeepConf@512 achieves up to 99.9% accuracy and reduces generated tokens by up to 84.7% compared to full parallel thinking.Figure 1: Up: DeepConf on AIME 2025. Work done during an internship at Meta FAIR. Large Language Models (LLMs) have demonstrated exceptional reasoning capabilities, particularly when equipped with methods that enhance their performance during test-time inference. A prominent technique is self-consistency, which samples multiple reasoning paths and aggregates final answers through majority voting (Wang et al., 2023). This type of approach, also known as parallel thinking, significantly improves reasoning accuracy but incurs substantial computational overhead: generating numerous reasoning traces per query scales inference overhead linearly, limiting practical deployment (Xue et al., 2023). For example, improving pass@1 accuracy from 68% to 82% using standard majority voting on AIME 2025 requires 511 additional reasoning traces per question using Qwen3-8B, consuming 100 million additional tokens. Moreover, parallel thinking with majority voting exhibits diminishing returns --performance often saturates or degrades as the number of traces increase (Chen et al., 2024a).

Automatic evaluation of generative tasks using large language models faces challenges due to ambiguous criteria. Although automatic checklist generation is a potentially promising approach, its usefulness remains underexplored. We investigate whether checklists should be used for all questions or selectively, generate them using six methods, evaluate their effectiveness across eight model sizes, and identify checklist items that correlate with human evaluations. Through experiments on pairwise comparison and direct scoring tasks, we find that selective checklist use tends to improve evaluation performance in pairwise settings, while its benefits are less consistent in direct scoring. Our analysis also shows that even checklist items with low correlation to human scores often reflect human-written criteria, indicating potential inconsistencies in human evaluation. These findings highlight the need to more clearly define objective evaluation criteria to guide both human and automatic evaluations. \footnote{Our code is available at~https://github.com/momo0817/checklist-effectiveness-study

However, the reliability of large language models (LLMs) when reasoning under high-constraint regimes is insufficiently characterized. To address this gap, we present R-ConstraintBench, a scalable framework that evaluates models on Resource-Constrained Project Scheduling Problems (RCPSP), an NP-Complete feasibility class, while difficulty increases via linear growth in constraints. R-ConstraintBench incrementally increases non-redundant precedence constraints in Directed Acyclic Graphs (DAGs) and then introduces downtime, temporal windows, and disjunctive constraints. As an illustrative example, we instantiate the benchmark in a data center migration setting and evaluate multiple LLMs using feasibility and error analysis, identifying degradation thresholds and constraint types most associated with failure. Empirically, strong models are near-ceiling on precedence-only DAGs, but feasibility performance collapses when downtime, temporal windows, and disjunctive constraints interact--implicating constraint interaction, not graph depth, as the principal bottleneck. Performance on clean synthetic ramps also does not guarantee transfer to domain-grounded scenarios, underscoring limited generalization.

Process Reward Models (PRMs) have emerged as a promising framework for supervising intermediate reasoning in large language models (LLMs), yet existing PRMs are primarily trained on general or Science, Technology, Engineering, and Mathematics (STEM) domains and fall short in domain-specific contexts such as finance, where reasoning is more structured, symbolic, and sensitive to factual and regulatory correctness. We introduce \textbf{Fin-PRM}, a domain-specialized, trajectory-aware PRM tailored to evaluate intermediate reasoning steps in financial tasks. Fin-PRM integrates step-level and trajectory-level reward supervision, enabling fine-grained evaluation of reasoning traces aligned with financial logic. We apply Fin-PRM in both offline and online reward learning settings, supporting three key applications: (i) selecting high-quality reasoning trajectories for distillation-based supervised fine-tuning, (ii) providing dense process-level rewards for reinforcement learning, and (iii) guiding reward-informed Best-of-N inference at test time. Experimental results on financial reasoning benchmarks, including CFLUE and FinQA, demonstrate that Fin-PRM consistently outperforms general-purpose PRMs and strong domain baselines in trajectory selection quality. Downstream models trained with Fin-PRM yield substantial improvements with baselines, with gains of 12.9\% in supervised learning, 5.2\% in reinforcement learning, and 5.1\% in test-time performance. These findings highlight the value of domain-specialized reward modeling for aligning LLMs with expert-level financial reasoning. Our project resources will be available at https://github.com/aliyun/qwen-dianjin.

While large language models (LLMs) have shown promise in healthcare, their application for rare medical conditions is still hindered by scarce and unreliable datasets for fine-tuning. Hyperhidrosis, a disorder causing excessive sweating beyond physiological needs, is one such rare disorder, affecting 2-3% of the population and significantly impacting both physical comfort and psychosocial well-being. To date, no work has tailored LLMs to advance the diagnosis or care of hyperhidrosis. To address this gap, we present LLM4Sweat, an open-source and domain-specific LLM framework for trustworthy and empathetic hyperhidrosis support. The system follows a three-stage pipeline. In the data augmentation stage, a frontier LLM generates medically plausible synthetic vignettes from curated open-source data to create a diverse and balanced question-answer dataset. In the fine-tuning stage, an open-source foundation model is fine-tuned on the dataset to provide diagnosis, personalized treatment recommendations, and empathetic psychological support. In the inference and expert evaluation stage, clinical and psychological specialists assess accuracy, appropriateness, and empathy, with validated responses iteratively enriching the dataset. Experiments show that LLM4Sweat outperforms baselines and delivers the first open-source LLM framework for hyperhidrosis, offering a generalizable approach for other rare diseases with similar data and trustworthiness challenges.

Tokenization plays a critical role in language modeling, yet existing approaches such as Byte-Pair Encoding (BPE) or WordPiece operate purely on frequency statistics, ignoring the underlying semantic structure of text. This leads to over-tokenization of semantically redundant spans and underutilization of contextual coherence, particularly in long-context scenarios. In this work, we propose \textbf{SemToken}, a semantic-aware tokenization framework that jointly reduces token redundancy and improves computation efficiency. SemToken first extracts contextual semantic embeddings via lightweight encoders and performs local semantic clustering to merge semantically equivalent tokens. Then, it allocates heterogeneous token granularity based on semantic density, allowing finer-grained tokenization in content-rich regions and coarser compression in repetitive or low-entropy spans. SemToken can be seamlessly integrated with modern language models and attention acceleration methods. Experiments on long-context language modeling benchmarks such as WikiText-103 and LongBench show that SemToken achieves up to $2.4\times$ reduction in token count and $1.9\times$ speedup, with negligible or no degradation in perplexity and downstream accuracy. Our findings suggest that semantic structure offers a promising new axis for optimizing tokenization and computation in large language models.

--Jailbreak attacks pose a serious threat to the safety of Large Language Models (LLMs) by crafting adversarial prompts that bypass alignment mechanisms, causing the models to produce harmful, restricted, or biased content. In this paper, we propose SafeLLM, a novel unlearning-based defense framework that unlearn the harmful knowledge from LLMs while preserving linguistic fluency and general capabilities. SafeLLM employs a three-stage pipeline: (1) dynamic unsafe output detection using a hybrid approach that integrates external classifiers with model-internal evaluations; (2) token-level harmful content tracing through feedforward network (FFN) activations to localize harmful knowledge; and (3) constrained optimization to suppress unsafe behavior without degrading overall model quality. SafeLLM achieves targeted and irreversible forgetting by identifying and neutralizing FFN substructures responsible for harmful generation pathways. Extensive experiments on prominent LLMs (Vicuna, LLaMA, and GPT -J) across multiple jailbreak benchmarks show that SafeLLM substantially reduces attack success rates while maintaining high general-purpose performance. Compared to standard defense methods such as supervised fine-tuning and direct preference optimization, SafeLLM offers stronger safety guarantees, more precise control over harmful behavior, and greater robustness to unseen attacks. Moreover, SafeLLM maintains the general performance after the harmful knowledge unlearned. Large Language Models (LLMs) are a class of foundation models trained on massive datasets, enabling them to understand and generate not only natural language but also a variety of other content types. These capabilities enable LLMs to perform a wide array of tasks, ranging from general-purpose language processing to domain-specific applications across fields such as healthcare, law, finance, and education. Built on deep learning architectures like Transformers, LLMs excel in tasks including summarization, translation, question answering, and sentiment analysis.

Existing unlearning methods often rely on the assumption of having access to the entire training dataset during the forgetting process. However, this assumption may not hold true in practical scenarios where the original training data may not be accessible, i.e., the source-free setting. T o address this challenge, we focus on the source-free unlearning scenario, where an unlearning algorithm must be capable of removing specific data from a trained model without requiring access to the original training dataset. Building on recent work, we present a method that can estimate the Hessian of the unknown remaining training data, a crucial component required for efficient unlearning. Leveraging this estimation technique, our method enables efficient zero-shot unlearning while providing robust theoretical guarantees on the unlearning performance, while maintaining performance on the remaining data. Extensive experiments over a wide range of datasets verify the efficacy of our method.