Papers

Multi-stage reasoning has emerged as an effective strategy for enhancing the reasoning capability of small language models by decomposing complex problems into sequential sub-stages. However, this comes at the cost of increased latency. We observe that existing adaptive acceleration techniques, such as layer skipping, struggle to balance efficiency and accuracy in this setting due to two key challenges: (1) stage-wise variation in skip sensitivity, and (2) the generation of redundant output tokens. To address these, we propose LiteStage, a latency-aware layer skipping framework for multi-stage reasoning. LiteStage combines a stage-wise offline search that allocates optimal layer budgets with an online confidence-based generation early exit to suppress unnecessary decoding. Experiments on three benchmarks, e.g., OBQA, CSQA, and StrategyQA, show that LiteStage outperforms prior training-free layer skipping methods.

The emerging large language model role-playing agents (LLM RPAs) aim to simulate individual human behaviors, but the persona fidelity is often undermined by manually-created profiles (e.g., cherry-picked information and personality characteristics) without validating the alignment with the target individuals. To address this limitation, our work introduces the Dynamic Persona Refinement Framework (DPRF). DPRF aims to optimize the alignment of LLM RPAs' behaviors with those of target individuals by iteratively identifying the cognitive divergence, either through free-form or theory-grounded, structured analysis, between generated behaviors and human ground truth, and refining the persona profile to mitigate these divergences. We evaluate DPRF with five LLMs on four diverse behavior-prediction scenarios: formal debates, social media posts with mental health issues, public interviews, and movie reviews. DPRF can consistently improve behavioral alignment considerably over baseline personas and generalizes across models and scenarios. Our work provides a robust methodology for creating high-fidelity persona profiles and enhancing the validity of downstream applications, such as user simulation, social studies, and personalized AI.

This paper proposes a joint modeling method of the Big Five, which has long been studied, and HEXACO, which has recently attracted attention in psychology, for automatically recognizing apparent personality traits from multimodal human behavior. Most previous studies have used the Big Five for multimodal apparent personality-trait recognition. However, no study has focused on apparent HEXACO which can evaluate an Honesty-Humility trait related to displaced aggression and vengefulness, social-dominance orientation, etc. In addition, the relationships between the Big Five and HEXACO when modeled by machine learning have not been clarified. We expect awareness of multimodal human behavior to improve by considering these relationships. The key advance of our proposed method is to optimize jointly recognizing the Big Five and HEXACO. Experiments using a self-introduction video dataset demonstrate that the proposed method can effectively recognize the Big Five and HEXACO.

After the pretraining stage of LLMs, techniques such as SFT, RLHF, RLVR, and RFT are applied to enhance instruction-following ability, mitigate undesired responses, improve reasoning capability and enable efficient domain adaptation with minimal data. SFT relies on the next-token prediction objective to strengthen instruction following in a base model using a large corpus of human-labeled responses. In contrast, RFT employs a RL-based approach to adapt fine-tuned reasoning models to specific domains with limited supervision. Inspired by RFT, we propose replacing SFT with RLSR to leverage the extensive SFT dataset in an RL framework, thereby improving the base model's instruction-following ability. In RLSR, the base model generates multiple responses for each prompt, and reward scores are computed as the cosine similarity in the semantic embedding space between the generated and human-labeled responses. RLSR can be utilized in multiple ways. It can directly replace SFT, achieving superior performance on instruction-following benchmarks-for example, RLSR (SB) on Qwen-7B (INFINITY) achieved an AlpacaEval win rate of 26.34%, surpassing SFT's 21.01%. Furthermore, combining SFT and RLSR further enhances downstream task performance; Qwen-7B (INFINITY) achieved a win rate of 30.73% when trained with SFT + RLSR.

We present MAFA (Multi-Agent Framework for Annotation), a production-deployed system that transforms enterprise-scale annotation workflows through configurable multi-agent collaboration. Addressing the critical challenge of annotation backlogs in financial services, where millions of customer utterances require accurate categorization, MAFA combines specialized agents with structured reasoning and a judge-based consensus mechanism. Our framework uniquely supports dynamic task adaptation, allowing organizations to define custom annotation types (FAQs, intents, entities, or domain-specific categories) through configuration rather than code changes. Deployed at JP Morgan Chase, MAFA has eliminated a 1 million utterance backlog while achieving, on average, 86% agreement with human annotators, annually saving over 5,000 hours of manual annotation work. The system processes utterances with annotation confidence classifications, which are typically 85% high, 10% medium, and 5% low across all datasets we tested. This enables human annotators to focus exclusively on ambiguous and low-coverage cases. We demonstrate MAFA's effectiveness across multiple datasets and languages, showing consistent improvements over traditional and single-agent annotation baselines: 13.8% higher Top-1 accuracy, 15.1% improvement in Top-5 accuracy, and 16.9% better F1 in our internal intent classification dataset and similar gains on public benchmarks. This work bridges the gap between theoretical multi-agent systems and practical enterprise deployment, providing a blueprint for organizations facing similar annotation challenges.

Recent Large Language Models (LLMs) have demonstrated remarkable proficiency in code generation. However, their ability to create complex visualizations for scaled and structured data remains largely unevaluated and underdeveloped. To address this gap, we introduce PlotCraft, a new benchmark featuring 1k challenging visualization tasks that cover a wide range of topics, such as finance, scientific research, and sociology. The benchmark is structured around seven high-level visualization tasks and encompasses 48 distinct chart types. Crucially, it is the first to systematically evaluate both single-turn generation and multi-turn refinement across a diverse spectrum of task complexities. Our comprehensive evaluation of 23 leading LLMs on PlotCraft reveals obvious performance deficiencies in handling sophisticated visualization tasks. To bridge this performance gap, we develope SynthVis-30K, a large-scale, high-quality dataset of complex visualization code synthesized via a collaborative agent framework. Building upon this dataset, we develope PlotCraftor, a novel code generation model that achieves strong capabilities in complex data visualization with a remarkably small size. Across VisEval, PandasPlotBench, and our proposed PlotCraft, PlotCraftor shows performance comparable to that of leading proprietary approaches. Especially, on hard task, Our model achieves over 50% performance improvement. We will release the benchmark, dataset, and code at https://github.com/Speakn0w/PlotCraft-Benchmark.

Infrastructure in smart cities is increasingly monitored by networks of closed circuit television (CCTV) cameras. Roads, bridges and tunnels develop cracks, potholes, and fluid leaks that threaten public safety and require timely repair. Manual inspection is costly and hazardous, and existing automatic systems typically address individual defect types or provide unstructured outputs that cannot directly guide maintenance crews. This paper proposes a comprehensive pipeline that leverages street CCTV streams for multi defect detection and segmentation using the YOLO family of object detectors and passes the detections to a vision language model (VLM) for scene aware summarization. The VLM generates a structured action plan in JSON format that includes incident descriptions, recommended tools, dimensions, repair plans, and urgent alerts. We review literature on pothole, crack and leak detection, highlight recent advances in large vision language models such as QwenVL and LLaVA, and describe the design of our early prototype. Experimental evaluation on public datasets and captured CCTV clips demonstrates that the system accurately identifies diverse defects and produces coherent summaries. We conclude by discussing challenges and directions for scaling the system to city wide deployments.

Model merging aims to combine multiple fine-tuned models into a single set of weights that performs well across all source tasks. While prior work has shown that merging can approximate the performance of individual fine-tuned models for each task, it largely overlooks scenarios where models are compressed into low-rank representations, either through low-rank adaptation (LoRA) or post-training singular value decomposition (SVD). We first demonstrate that applying conventional merging methods to low-rank weights leads to severe performance degradation in the merged model. Motivated by this phenomenon, we propose a fundamentally different approach: instead of collapsing all adapters into one set of weights, we construct a compact basis (e.g., an equivalent of holding two or more models) from which original task-specific models can be recovered via linear combination. This reframes merging as generating a reconstruction-capable model space rather than producing a single merged model. Crucially, this allows us to “revert” to each individual model when needed, recognizing that no merged model can consistently outperform one specialized for its task. Building on this insight, we introduce our method, Reversible Model Merging (RMM), an efficient, data-free, and flexible method that provides a closed-form solution for selecting the optimal basis of model weights and task-specific coefficients for linear combination. Extensive experiments across diverse datasets and model scales demonstrate that RMM consistently outperforms existing merging approaches, preserving the performance of low-rank compressed models by a significant margin.

Computational gastronomy increasingly relies on diverse, high-quality recipe datasets to capture regional culinary traditions. Although there are large-scale collections for major languages, Macedonian recipes remain under-represented in digital research. In this work, we present the first systematic effort to construct a Macedonian recipe dataset through web scraping and structured parsing. We address challenges in processing heterogeneous ingredient descriptions, including unit, quantity, and descriptor normalization. An exploratory analysis of ingredient frequency and co-occurrence patterns, using measures such as Pointwise Mutual Information and Lift score, highlights distinctive ingredient combinations that characterize Macedonian cuisine. The resulting dataset contributes a new resource for studying food culture in underrepresented languages and offers insights into the unique patterns of Macedonian culinary tradition.

Large language models (LLMs) are transforming everyday applications, yet deployment in cybersecurity lags due to a lack of high-quality, domain-specific models and training datasets. To address this gap, we present CyberPal 2.0, a family of cybersecurity-expert small language models (SLMs) ranging from 4B-20B parameters. To train CyberPal 2.0, we generate an enriched chain-of-thought cybersecurity instruction dataset built with our data enrichment and formatting pipeline, SecKnowledge 2.0, which integrates expert-in-the-loop steering of reasoning formats alongside LLM-driven multi-step grounding, yielding higher-fidelity, task-grounded reasoning traces for security tasks. Across diverse cybersecurity benchmarks, CyberPal 2.0 consistently outperforms its baselines and matches or surpasses various open and closed-source frontier models, while remaining a fraction of their size. On core cyber threat intelligence knowledge tasks, our models outperform almost all tested frontier models, ranking second only to Sec-Gemini v1. On core threat-investigation tasks, such as correlating vulnerabilities and bug tickets with weaknesses, our best 20B-parameter model outperforms GPT-4o, o1, o3-mini, and Sec-Gemini v1, ranking first, while our smallest 4B-parameter model ranks second.

Detecting out-of-scope (OOS) user utterances remains a key challenge in task-oriented dialogue systems and, more broadly, in open-set intent recognition. Existing approaches often depend on strong distributional assumptions or auxiliary calibration modules. We present DROID (Dual Representation for Out-of-Scope Intent Detection), a compact end-to-end framework that combines two complementary encoders – the Universal Sentence Encoder (USE) for broad semantic generalization and a domain-adapted Transformer-based Denoising Autoencoder (TSDAE) for domain-specific contextual distinctions. Their fused representations are processed by a lightweight branched classifier with a single calibrated threshold that separates in-domain and OOS intents without post-hoc scoring. To enhance boundary learning under limited supervision, DROID incorporates both synthetic and open-domain outlier augmentation. Despite using only 1.5M trainable parameters, DROID consistently outperforms recent state-of-the-art baselines across multiple intent benchmarks, achieving macro-F1 improvements of 6–15% for known and 8–20% for OOS intents, with the most significant gains in low-resource settings. These results demonstrate that dual-encoder representations with simple calibration can yield robust, scalable, and reliable OOS detection for neural dialogue systems.

Current frameworks for consensus statement generation with large language models lack the inherent structure needed to provide provable fairness guarantees when aggregating diverse free-form opinions. We model the task as a multi-objective, token-level Markov Decision Process (MDP), where each objective corresponds to an agent's preference. Token-level rewards for each agent are derived from their policy (e.g., a personalized language model). This approach utilizes the finding that such policies implicitly define optimal Q-functions, providing a principled way to quantify rewards at each generation step without a value function (Rafailov et al., 2024). This MDP formulation creates a formal structure amenable to analysis using principles from social choice theory. We propose two approaches grounded in social choice theory. First, we propose a stochastic generation policy guaranteed to be in the ex-ante core, extending core stability concepts from voting theory to text generation. This policy is derived from an underlying distribution over complete statements that maximizes proportional fairness (Nash Welfare). Second, for generating a single statement, we target the maximization of egalitarian welfare using search algorithms within the MDP framework. Empirically, experiments using language models to instantiate agent policies show that search guided by the egalitarian objective generates consensus statements with improved worst-case agent alignment compared to baseline methods, including the Habermas Machine (Tessler et al., 2024).

Large Language Models (LLMs) suffer significant performance degradation in multi-turn conversations when information is presented incrementally. Given that multi-turn conversations characterize everyday interactions with LLMs, this degradation poses a severe challenge to real world usability. We hypothesize that abrupt increases in model uncertainty signal misalignment in multi-turn LLM interactions, and we exploit this insight to dynamically realign conversational context. We introduce ERGO (Entropy-guided Resetting for Generation Optimization), which continuously quantifies internal uncertainty via Shannon entropy over next token distributions and triggers adaptive prompt consolidation when a sharp spike in entropy is detected. By treating uncertainty as a first class signal rather than a nuisance to eliminate, ERGO embraces variability in language and modeling, representing and responding to uncertainty. In multi-turn tasks with incrementally revealed instructions, ERGO yields a 56.6% average performance gain over standard baselines, increases aptitude (peak performance capability) by 24.7%, and decreases unreliability (variability in performance) by 35.3%, demonstrating that uncertainty aware interventions can improve both accuracy and reliability in conversational AI.

Language is, as commonly theorized, largely arbitrary. Yet, systematic relationships between phonetics and semantics have been observed in many specific cases. To what degree could those systematic relationships manifest themselves in large scale, quantitative investigations–both in previously identified and unidentified phenomena? This work undertakes a distributional approach to quantifying phonosemantic iconicity at scale across 6 diverse languages (English, Spanish, Hindi, Finnish, Turkish, and Tamil). In each language, we analyze the alignment of morphemes' phonetic and semantic similarity spaces with a suite of statistical measures, and discover an array of interpretable phonosemantic alignments not previously identified in the literature, along with crosslinguistic patterns. We also analyze 5 previously hypothesized phonosemantic alignments, finding support for some such alignments and mixed results for others.

Large language models (LLMs) can exhibit biases in reasoning capabilities due to linguistic modality, performing better on tasks in one language versus another, even with similar content. Most previous works evaluate this through reasoning tasks where reliance on strategies or knowledge can ensure success, such as in commonsense or math tasks. However, abstract reasoning is vital to reasoning for everyday life, where people apply "out-of-the-box thinking" to identify and use patterns for solutions, without a reliance on formulaic approaches. Comparatively, little work has evaluated linguistic biases in this task type. In this paper, we propose a task inspired by the New York Times Connections: GlobalGroup, that evaluates models in an abstract reasoning task across several languages. We constructed a game benchmark with five linguistic backgrounds – English, Spanish, Chinese, Hindi, and Arabic – in both the native language and an English translation for comparison. We also proposed game difficulty measurements to evaluate models on games with similar difficulty, enabling a more controlled comparison, which is particularly important in reasoning evaluations. Through experimentation, we find English modalities largely lead to better performance in this abstract reasoning task, and performance disparities between open- and closed-source models.

Correct answers do not necessarily reflect cultural understanding. We introduce CRaFT, an explanation-based multilingual evaluation framework designed to assess how large language models (LLMs) reason across cultural contexts. Rather than scoring outputs solely based on accuracy, CRaFT evaluates model explanations using four interpretable metrics: Cultural Fluency, Deviation, Consistency, and Linguistic Adaptation. We apply the framework to 50 culturally grounded questions from the World Values Survey, translated into Arabic, Bengali, and Spanish, and evaluate three models (GPT, DeepSeek, and FANAR) across over 2,100 answer-explanation pairs. Results reveal significant cross-lingual variation in reasoning: Arabic reduces fluency, Bengali enhances it, and Spanish remains largely stable. While GPT adapts more effectively across languages, it exhibits lower consistency; FANAR shows stable but rigid reasoning. These findings suggest that cultural awareness in LLMs is not intrinsic but emerges through linguistic framing. CRaFT offers a new lens for evaluating cross-cultural reasoning in multilingual settings, providing actionable insights for building culturally adaptive language models.

In this paper, we present BitNet Distillation (BitDistill), a lightweight pipeline that fine-tunes off-the-shelf full-precision LLMs (e.g., Qwen) into 1.58-bit precision (i.e., ternary weights {-1, 0, 1}) for specific downstream tasks, achieving strong task-specific performance with minimal computational cost. Specifically, BitDistill incorporates three key techniques: the SubLN module, as introduced in BitNet; multi-head attention distillation, based on MiniLM; and continual pre-training, which serves as a crucial warm-up step to mitigate the scalability issue of the performance gap between finetuned full-precision and 1.58-bit LLMs on specific tasks. Experimental results show that BitDistill achieves performance comparable to the full-precision counterpart models across model size, while enabling up to 10x memory savings and 2.65x faster inference on CPUs. Code is available at https://github.com/microsoft/BitNet.

Large language model development relies on large-scale training corpora, yet most contain data of unclear licensing status, limiting the development of truly open models. This problem is exacerbated for non-English languages, where openly licensed text remains critically scarce. We introduce the German Commons, the largest collection of openly licensed German text to date. It compiles data from 41 sources across seven domains, encompassing legal, scientific, cultural, political, news, economic, and web text. Through systematic sourcing from established data providers with verifiable licensing, it yields 154.56 billion tokens of high-quality text for language model training. Our processing pipeline implements comprehensive quality filtering, deduplication, and text formatting fixes, ensuring consistent quality across heterogeneous text sources. All domain subsets feature licenses of at least CC-BY-SA 4.0 or equivalent, ensuring legal compliance for model training and redistribution. The German Commons therefore addresses the critical gap in openly licensed German pretraining data, and enables the development of truly open German language models. We also release code for corpus construction and data filtering tailored to German language text, rendering the German Commons fully reproducible and extensible.

State-of-the-art (SOTA) Automatic Speech Recognition (ASR) systems primarily rely on acoustic information while disregarding additional multi-modal context. However, visual information are essential in disambiguation and adaptation. While most work focus on speaker images to handle noise conditions, this work also focuses on integrating presentation slides for the use cases of scientific presentation. In a first step, we create a benchmark for multi-modal presentation including an automatic analysis of transcribing domain-specific terminology. Next, we explore methods for augmenting speech models with multi-modal information. We mitigate the lack of datasets with accompanying slides by a suitable approach of data augmentation. Finally, we train a model using the augmented dataset, resulting in a relative reduction in word error rate of approximately 34%, across all words and 35%, for domain-specific terms compared to the baseline model.

Retrieval-augmented generation (RAG) is a prevalent approach for building LLM-based question-answering systems that can take advantage of external knowledge databases. Due to the complexity of real-world RAG systems, there are many potential causes for erroneous outputs. Understanding the range of errors that can occur in practice is crucial for robust deployment. We present a new taxonomy of the error types that can occur in realistic RAG systems, examples of each, and practical advice for addressing them. Additionally, we curate a dataset of erroneous RAG responses annotated by error types. We then propose an auto-evaluation method aligned with our taxonomy that can be used in practice to track and address errors during development. Code and data are available at https://github.com/layer6ai-labs/rag-error-classification.

Recent progress in large language models (LLMs) has focused on test-time scaling to improve reasoning via increased inference computation, but often at the cost of efficiency. We revisit test-time behavior and uncover a simple yet underexplored phenomenon: reasoning uncertainty is highly localized-only a small subset of high-entropy tokens dominantly affects output correctness. Motivated by this, we propose Minimal Test-Time Intervention (MTI), a training-free framework that enhances reasoning accuracy and stability with minimal overhead. MTI includes: (i) Selective CFG intervention, applying classifier-free guidance only at uncertain positions; and (ii) Lightweight negative-prompt guidance, reusing the main model's KV cache to approximate unconditional decoding efficiently. MTI yields consistent gains across general, coding, and STEM tasks-e.g., +9.28% average improvement on six benchmarks for DeepSeek-R1-7B and +11.25% on AIME2024 using Ling-mini-2.0-while remaining highly efficient.

The use of copyrighted books for training AI has sparked lawsuits from authors concerned about AI generating derivative content. Yet whether these models can produce high-quality literary text emulating authors' voices remains unclear. We conducted a preregistered study comparing MFA-trained writers with three frontier models (ChatGPT, Claude, Gemini) writing up to 450-word excerpts emulating 50 award-winning authors' styles. In blind pairwise evaluations by 28 MFA-trained readers and 516 college-educated general readers, AI text from in-context prompting was strongly disfavored by MFA readers for stylistic fidelity (OR=0.16) and quality (OR=0.13), while general readers showed no fidelity preference (OR=1.06) but favored AI for quality (OR=1.82). Fine-tuning ChatGPT on authors' complete works reversed these results: MFA readers favored AI for fidelity (OR=8.16) and quality (OR=1.87), with general readers showing even stronger preference (fidelity OR=16.65; quality OR=5.42). Both groups preferred fine-tuned AI, but the writer-type X reader-type interaction remained significant (p=0.021 for fidelity; p<10^-4 for quality), indicating general readers favored AI by a wider margin. Effects are robust under cluster-robust inference and generalize across authors in heterogeneity analyses. Fine-tuned outputs were rarely flagged as AI-generated (3% vs. 97% for prompting) by leading detectors. Mediation analysis shows fine-tuning eliminates detectable AI quirks that penalize in-context outputs, altering the nexus between detectability and preference. While not accounting for effort to transform AI output into publishable prose, the median fine-tuning cost of $81 per author represents a 99.7% reduction versus typical writer compensation. Author-specific fine-tuning enables non-verbatim AI writing preferred over expert human writing, providing evidence relevant to copyright's fourth fair-use factor.

Deep Research (DR) agents, powered by advanced Large Language Models (LLMs), have recently garnered increasing attention for their capability in conducting complex research tasks. However, existing literature lacks a rigorous and systematic evaluation of DR Agent's capabilities in critical research analysis. To address this gap, we first propose HisRubric, a novel evaluation framework with a hierarchical analytical structure and a fine-grained grading rubric for rigorously assessing DR agents' capabilities in corporate financial analysis. This framework mirrors the professional analyst's workflow, progressing from data recognition to metric calculation, and finally to strategic summarization and interpretation. Built on this framework, we construct a FinDeepResearch benchmark that comprises 64 listed companies from 8 financial markets across 4 languages, encompassing a total of 15,808 grading items. We further conduct extensive experiments on the FinDeepResearch using 16 representative methods, including 6 DR agents, 5 LLMs equipped with both deep reasoning and search capabilities, and 5 LLMs with deep reasoning capabilities only. The results reveal the strengths and limitations of these approaches across diverse capabilities, financial markets, and languages, offering valuable insights for future research and development. The benchmark and evaluation code is publicly available at https://OpenFinArena.com/.

The facts a language model stores are tied to its parameter count, so small models that fit on edge devices fail on expert problems, which need specialized knowledge and follow multi-step procedures. Fine-tuning for a specific domain or task writes the knowledge into the parameters but must be repeated for every model and domain, and a retrieved passage leaves the model to find the relevant fact and apply it on its own. We introduce instruction retrieval, which distills a teacher model's expertise into a corpus of instructions tailored so that a small model can follow. For each cluster of a domain's problems, the teacher writes one instruction with the background knowledge the cluster depends on, a procedure for that kind of problem, and the common mistakes made on it. This reusable corpus needs only to be built once per domain and requires no run-time teacher access. At inference, a frozen small model retrieves the instructions nearest its question and follows them, with no fine-tuning. Across medicine, law, and mathematics benchmarks, we demonstrate the corpus improves over zero-shot on every task and over few-shot prompting, self-consistency, and other retrieved text on medicine and law. On MedQA the corpus raises mean accuracy by 10.6 points, where retrieved textbook passages raise it by 3.9 and few-shot examples from the same teacher lower it. An error analysis shows that only the background knowledge fixes the questions a small model always gets wrong, and that the procedure and common mistakes fix only the questions where it wavers between options. Our results show that automatically retrieved inference-time procedural guidance and domain knowledge can yield substantial gains for small models.

Large language models (LLMs) are increasingly applied in biomedical domains, yet their reliability in drug-safety prediction remains underexplored. In this work, we investigate whether LLMs incorporate socio-demographic information into adverse event (AE) predictions, despite such attributes being clinically irrelevant. Using structured data from the United States Food and Drug Administration Adverse Event Reporting System (FAERS) and a persona-based evaluation framework, we assess two state-of-the-art models, ChatGPT-4o and Bio-Medical-Llama-3.8B, across diverse personas defined by education, marital status, employment, insurance, language, housing stability, and religion. We further evaluate performance across three user roles (general practitioner, specialist, patient) to reflect real-world deployment scenarios where commercial systems often differentiate access by user type. Our results reveal systematic disparities in AE prediction accuracy. Disadvantaged groups (e.g., low education, unstable housing) were frequently assigned higher predicted AE likelihoods than more privileged groups (e.g., postgraduate-educated, privately insured). Beyond outcome disparities, we identify two distinct modes of bias: explicit bias, where incorrect predictions directly reference persona attributes in reasoning traces, and implicit bias, where predictions are inconsistent, yet personas are not explicitly mentioned. These findings expose critical risks in applying LLMs to pharmacovigilance and highlight the urgent need for fairness-aware evaluation protocols and mitigation strategies before clinical deployment.

We propose and test the LLM Brain Rot Hypothesis: continual exposure to junk web text induces lasting cognitive decline in large language models (LLMs). To unveil junk effects, we designed a novel controlled experiment on real Twitter/X corpora, by constructing junk and reverse-controlled datasets via two orthogonal operationalizations: M1 (engagement degree) and M2 (semantic quality), with matched token scale and training operations across conditions. Compared to the control group, continual pre-training of 4 LLMs on the junk dataset causes non-trivial declines (Hedges' g>0.3) on reasoning, long-context understanding, safety, and inflating "dark traits" (e.g., psychopathy, narcissism). The gradual mixtures of junk and control datasets also yield dose-response cognition decay: for example, under M1, ARC-Challenge with Chain-of-Thought drops 72.1 -> 57.2 and RULER-CWE 83.7 -> 52.3 as junk ratio rises from 0% to 100%. Error forensics reveal several key insights. First, we identify thought-skipping as the primary lesion in reasoning: models increasingly truncate or skip chains. Second, partial but incomplete healing is observed: scaling instruction tuning and clean continual pre-training improve the declined cognition, yet cannot restore baseline capability, suggesting persistent representational drift rather than format mismatch. Finally, we discover that the popularity, a non-semantic metric, of a tweet is a better indicator of the Brain Rot effect than the length in M1. Together, the results provide significant, multi-perspective evidence that social effects of data could be a causal driver of LLM capability decay in continual pre-training, thereby motivating routine "cognitive health checks" for deployed and evolving LLMs.

Biomedical queries often rely on a deep understanding of specialized knowledge such as gene regulatory mechanisms and pathological processes of diseases. They require detailed analysis of complex physiological processes and effective integration of information from multiple data sources to support accurate retrieval and reasoning. Although large language models (LLMs) perform well in general reasoning tasks, their generated biomedical content often lacks scientific rigor due to the inability to access authoritative biomedical databases and frequently fabricates protein functions, interactions, and structural details that deviate from authentic information. Therefore, we present BioMedSearch, a multi-source biomedical information retrieval framework based on LLMs. The method integrates literature retrieval, protein database and web search access to support accurate and efficient handling of complex biomedical queries. Through sub-queries decomposition, keywords extraction, task graph construction, and multi-source information filtering, BioMedSearch generates high-quality question-answering results. To evaluate the accuracy of question answering, we constructed a multi-level dataset, BioMedMCQs, consisting of 3,000 questions. The dataset covers three levels of reasoning: mechanistic identification, non-adjacent semantic integration, and temporal causal reasoning, and is used to assess the performance of BioMedSearch and other methods on complex QA tasks. Experimental results demonstrate that BioMedSearch consistently improves accuracy over all baseline models across all levels. Specifically, at Level 1, the average accuracy increases from 59.1% to 91.9%; at Level 2, it rises from 47.0% to 81.0%; and at the most challenging Level 3, the average accuracy improves from 36.3% to 73.4%. The code and BioMedMCQs are available at: https://github.com/CyL-ucas/BioMed_Search

Internet of Things (IoT) networks generate diverse and high-volume traffic that reflects both normal activity and potential threats. Deriving meaningful insight from such telemetry requires cross-layer interpretation of behaviors, protocols, and context rather than isolated detection. This work presents an LLM-powered AI agent framework that converts raw packet captures into structured and semantically enriched representations for interactive analysis. The framework integrates feature extraction, transformer-based anomaly detection, packet and flow summarization, threat intelligence enrichment, and retrieval-augmented question answering. An AI agent guided by a large language model performs reasoning over the indexed traffic artifacts, assembling evidence to produce accurate and human-readable interpretations. Experimental evaluation on multiple IoT captures and six open models shows that hybrid retrieval, which combines lexical and semantic search with reranking, substantially improves BLEU, ROUGE, METEOR, and BERTScore results compared with dense-only retrieval. System profiling further indicates low CPU, GPU, and memory overhead, demonstrating that the framework achieves holistic and efficient interpretation of IoT network traffic.

Clinical notes contain unstructured text provided by clinicians during patient encounters. These notes are usually accompanied by a sequence of diagnostic codes following the International Classification of Diseases (ICD). Correctly assigning and ordering ICD codes is essential for medical diagnosis and reimbursement. However, automating this task remains challenging. State-of-the-art methods treated this problem as a classification task, leading to ignoring the order of ICD codes that is essential for different purposes. In this work, as a first attempt, we approach this task from a retrieval system perspective to consider the order of codes, thus formulating this problem as a classification and ranking task. Our results and analysis show that the proposed framework has a superior ability to identify high-priority codes compared to other methods. For instance, our model's accuracy in correctly ranking primary diagnosis codes is 47%, compared to 20% for the state-of-the-art classifier. Additionally, in terms of classification metrics, the proposed model achieves a micro- and macro-F1 scores of 0.6065 and 0.2904, respectively, surpassing the previous best model with scores of 0.6035 and 0.2741.

Query-focused table summarization requires generating natural language summaries of tabular data conditioned on a user query, enabling users to access insights beyond fact retrieval. Existing approaches face key limitations: table-to-text models require costly fine-tuning and struggle with complex reasoning, prompt-based LLM methods suffer from token-limit and efficiency issues while exposing sensitive data, and prior agentic pipelines often rely on decomposition, planning, or manual templates that lack robustness and scalability. To mitigate these issues, we introduce an agentic workflow, FACTS, a Fast, Accurate, and Privacy-Compliant Table Summarization approach via Offline Template Generation. FACTS produces offline templates, consisting of SQL queries and Jinja2 templates, which can be rendered into natural language summaries and are reusable across multiple tables sharing the same schema. It enables fast summarization through reusable offline templates, accurate outputs with executable SQL queries, and privacy compliance by sending only table schemas to LLMs. Evaluations on widely-used benchmarks show that FACTS consistently outperforms baseline methods, establishing it as a practical solution for real-world query-focused table summarization. Our code is available at https://github.com/BorealisAI/FACTS.