Large Language Models (LLMs) exhibit impressive capabilities, but remain susceptible to a growing spectrum of safety risks, including jailbreaks, toxic content, hallucinations, and bias. Existing defenses often address only a single threat type or resort to rigid outright rejection, sacrificing user experience and failing to generalize across diverse and novel attacks. This paper introduces Adversarial Scenario Extrapolation (ASE), a novel inference-time computation framework that leverages Chain-of-Thought (CoT) reasoning to simultaneously enhance LLM robustness and seamlessness. ASE guides the LLM through a self-generative process of contemplating potential adversarial scenarios and formulating defensive strategies before generating a response to the user query. Comprehensive evaluation on four adversarial benchmarks with four latest LLMs shows that ASE achieves near-zero jailbreak attack success rates and minimal toxicity, while slashing outright rejections to <4%. ASE outperforms six state-of-the-art defenses in robustness-seamlessness trade-offs, with 92-99% accuracy on adversarial Q&A and 4-10x lower bias scores. By transforming adversarial perception into an intrinsic cognitive process, ASE sets a new paradigm for secure and natural human-AI interaction.

Large Reasoning Models (LRMs) introduce a new generation paradigm of explicitly reasoning before answering, leading to remarkable improvements in complex tasks. However, they pose great safety risks against harmful queries and adversarial attacks. While recent mainstream safety efforts on LRMs, supervised fine-tuning (SFT), improve safety performance, we find that SFT-aligned models struggle to generalize to unseen jailbreak prompts. After thorough investigation of LRMs' generation, we identify a safety aha moment that can activate safety reasoning and lead to a safe response. This aha moment typically appears in the `key sentence', which follows models' query understanding process and can indicate whether the model will proceed safely. Based on these insights, we propose SafeKey, including two complementary objectives to better activate the safety aha moment in the key sentence: (1) a Dual-Path Safety Head to enhance the safety signal in the model's internal representations before the key sentence, and (2) a Query-Mask Modeling objective to improve the models' attention on its query understanding, which has important safety hints. Experiments across multiple safety benchmarks demonstrate that our methods significantly improve safety generalization to a wide range of jailbreak attacks and out-of-distribution harmful prompts, lowering the average harmfulness rate by 9.6\%, while maintaining general abilities. Our analysis reveals how SafeKey enhances safety by reshaping internal attention and improving the quality of hidden representations.

The rapid advancement of language models has demonstrated the potential of artificial intelligence in the healthcare industry. However, small language models struggle with specialized domains in low-resource languages like Persian. While numerous medical-domain websites exist in Persian, no curated dataset or corpus has been available making ours the first of its kind. This study introduces a newly curated dataset comprising 20k doctor-patient Q\&A pairs and 60\% of a 90-million-token crawled corpus from medical magazines. Using a parameter-efficient fine-tuning approach, we enhanced the medical knowledge of the baseline model, aya-expanse-8b. Benchmark evaluations demonstrate that the fine-tuned model achieves improved accuracy in medical question answering and successfully passed the Iranian Basic Medical Science Entrance Exam (IBSEE) in September 2023, which the baseline model did not. Additionally, the fine-tuned model improved Persian-translated MMLU accuracy by an average of 2.67\%. This work highlights the potential of leveraging open-access online data to enrich small language models in medical fields, providing a novel solution for Persian medical AI applications suitable for resource-constrained environments. Future research could explore multimodal input to further enhance performance.

Recently, large vision-language models (LVLMs) have emerged as the preferred tools for judging text-image alignment, yet their robustness along the visual modality remains underexplored. This work is the first study to address a key research question: Can adversarial visual manipulations systematically fool LVLM judges into assigning unfairly inflated scores? We define potential image induced biases within the context of T2I evaluation and examine how these biases affect the evaluations of LVLM judges. Moreover, we introduce a novel, fine-grained, multi-domain meta-evaluation benchmark named FRAME, which is deliberately constructed to exhibit diverse score distributions. By introducing the defined biases into the benchmark, we reveal that all tested LVLM judges exhibit vulnerability across all domains, consistently inflating scores for manipulated images. Further analysis reveals that combining multiple biases amplifies their effects, and pairwise evaluations are similarly susceptible. Moreover, we observe that visual biases persist under prompt-based mitigation strategies, highlighting the vulnerability of current LVLM evaluation systems and underscoring the urgent need for more robust LVLM judges.

Knowledge Base Question Answering (KBQA) aims to answer natural language questions using structured knowledge from KBs. While LLM-only approaches offer generalization, they suffer from outdated knowledge, hallucinations, and lack of transparency. Chain-based KG-RAG methods address these issues by incorporating external KBs, but are limited to simple chain-structured questions due to the absence of planning and logical structuring. Inspired by semantic parsing methods, we propose PDRR: a four-stage framework consisting of Predict, Decompose, Retrieve, and Reason. Our method first predicts the question type and decomposes the question into structured triples. Then retrieves relevant information from KBs and guides the LLM as an agent to reason over and complete the decomposed triples. Experimental results demonstrate that PDRR consistently outperforms existing methods across various LLM backbones and achieves superior performance on both chain-structured and non-chain complex questions.

While scaling the length of responses at test-time has been shown to markedly improve the reasoning abilities and performance of large language models (LLMs), it often results in verbose outputs and increases inference cost. Prior approaches for efficient test-time scaling, typically using universal budget constraints or query-level length optimization, do not leverage historical information from previous encounters with the same problem during training. We hypothesize that this limits their ability to progressively make solutions more concise over time. To address this, we present History-Aware Policy Optimization (HAPO), which keeps track of a history state (e.g., the minimum length over previously generated correct responses) for each problem. HAPO employs a novel length reward function based on this history state to incentivize the discovery of correct solutions that are more concise than those previously found. Crucially, this reward structure avoids overly penalizing shorter incorrect responses with the goal of facilitating exploration towards more efficient solutions. By combining this length reward with a correctness reward, HAPO jointly optimizes for correctness and efficiency. We use HAPO to train DeepSeek-R1-Distill-Qwen-1.5B, DeepScaleR-1.5B-Preview, and Qwen-2.5-1.5B-Instruct, and evaluate HAPO on several math benchmarks that span various difficulty levels. Experiment results demonstrate that HAPO effectively induces LLMs' concise reasoning abilities, producing length reductions of 33-59% with accuracy drops of only 2-5%.

The automation of scientific research through large language models (LLMs) presents significant opportunities but faces critical challenges in knowledge synthesis and quality assurance. We introduce Feedback-Refined Agent Methodology (FRAME), a novel framework that enhances medical paper generation through iterative refinement and structured feedback. Our approach comprises three key innovations: (1) A structured dataset construction method that decomposes 4,287 medical papers into essential research components through iterative refinement; (2) A tripartite architecture integrating Generator, Evaluator, and Reflector agents that progressively improve content quality through metric-driven feedback; and (3) A comprehensive evaluation framework that combines statistical metrics with human-grounded benchmarks. Experimental results demonstrate FRAME's effectiveness, achieving significant improvements over conventional approaches across multiple models (9.91% average gain with DeepSeek V3, comparable improvements with GPT-4o Mini) and evaluation dimensions. Human evaluation confirms that FRAME-generated papers achieve quality comparable to human-authored works, with particular strength in synthesizing future research directions. The results demonstrated our work could efficiently assist medical research by building a robust foundation for automated medical research paper generation while maintaining rigorous academic standards.

Objective: To develop and evaluate a scalable methodology for harmonizing inconsistent units in large-scale clinical datasets, addressing a key barrier to data interoperability. Materials and Methods: We designed a novel unit harmonization system combining BM25, sentence embeddings, Bayesian optimization, and a bidirectional transformer based binary classifier for retrieving and matching laboratory test entries. The system was evaluated using the Optum Clinformatics Datamart dataset (7.5 billion entries). We implemented a multi-stage pipeline: filtering, identification, harmonization proposal generation, automated re-ranking, and manual validation. Performance was assessed using Mean Reciprocal Rank (MRR) and other standard information retrieval metrics. Results: Our hybrid retrieval approach combining BM25 and sentence embeddings (MRR: 0.8833) significantly outperformed both lexical-only (MRR: 0.7985) and embedding-only (MRR: 0.5277) approaches. The transformer-based reranker further improved performance (absolute MRR improvement: 0.10), bringing the final system MRR to 0.9833. The system achieved 83.39\% precision at rank 1 and 94.66\% recall at rank 5. Discussion: The hybrid architecture effectively leverages the complementary strengths of lexical and semantic approaches. The reranker addresses cases where initial retrieval components make errors due to complex semantic relationships in medical terminology. Conclusion: Our framework provides an efficient, scalable solution for unit harmonization in clinical datasets, reducing manual effort while improving accuracy. Once harmonized, data can be reused seamlessly in different analyses, ensuring consistency across healthcare systems and enabling more reliable multi-institutional studies and meta-analyses.

We present Distributional RewArds for Generative OptimizatioN (DRAGON), a versatile framework for fine-tuning media generation models towards a desired outcome. Compared with traditional reinforcement learning with human feedback (RLHF) or pairwise preference approaches such as direct preference optimization (DPO), DRAGON is more flexible. It can optimize reward functions that evaluate either individual examples or distributions of them, making it compatible with a broad spectrum of instance-wise, instance-to-distribution, and distribution-to-distribution rewards. Leveraging this versatility, we construct novel reward functions by selecting an encoder and a set of reference examples to create an exemplar distribution. When cross-modal encoders such as CLAP are used, the reference may be of a different modality (text versus audio). Then, DRAGON gathers online and on-policy generations, scores them with the reward function to construct a positive demonstration set and a negative set, and leverages the contrast between the two finite sets to approximate distributional reward optimization. For evaluation, we fine-tune an audio-domain text-to-music diffusion model with 20 reward functions, including a custom music aesthetics model, CLAP score, Vendi diversity, and Frechet audio distance (FAD). We further compare instance-wise (per-song) and full-dataset FAD settings while ablating multiple FAD encoders and reference sets. Over all 20 target rewards, DRAGON achieves an 81.45% average win rate. Moreover, reward functions based on exemplar sets enhance generations and are comparable to model-based rewards. With an appropriate exemplar set, DRAGON achieves a 60.95% human-voted music quality win rate without training on human preference annotations. DRAGON is a new approach to designing and optimizing reward functions for improving human-perceived quality. Demos at https://ml-dragon.github.io/web

The proliferation of large language models (LLMs) has significantly advanced intelligent systems. Unfortunately, LLMs often face knowledge conflicts between internal memory and retrieved external information, arising from misinformation, biases, or outdated knowledge. These conflicts undermine response reliability and introduce uncertainty in decision-making. In this work, we analyze how LLMs navigate knowledge conflicts from an information-theoretic perspective and reveal that when conflicting and supplementary information exhibit significant differences, LLMs confidently resolve their preferences and alleviate the uncertainty during their response generation. When this difference is ambiguous, LLMs experience considerable uncertainty about their generation. Based on this insight, we propose Swin-VIB, a novel framework that integrates a pipeline of variational information bottleneck models to adapt the retrieved information difference, facilitating robust response generation of LLMs even in conflicting contexts. Extensive experiments confirm our theoretical analysis and demonstrate the performance of Swin-VIB. Notably, Swin-VIB outperforms all competitive baselines in terms of the accuracy of the multiple-choice task, while improving the EM values in the open-ended QA task by at least 11.14%.