arXiv:2510.25908v1 Announce Type: new Abstract: Large language models (LLMs) have demonstrated transformative potential in scientific research, yet their deployment in high-stakes contexts raises significant trustworthiness concerns. Here, we introduce SciTrust 2.0, a comprehensive framework for evaluating LLM trustworthiness in scientific applications across four dimensions: truthfulness, adversarial robustness, scientific safety, and scientific ethics. Our framework incorporates novel, open-ended truthfulness benchmarks developed through a verified reflection-tuning pipeline and expert validation, alongside a novel ethics benchmark for scientific research contexts covering eight subcategories including dual-use research and bias. We evaluated seven prominent LLMs, including four science-specialized models and three general-purpose industry models, using multiple evaluation metrics including accuracy, semantic similarity measures, and LLM-based scoring. General-purpose industry models overall outperformed science-specialized models across each trustworthiness dimension, with GPT-o4-mini demonstrating superior performance in truthfulness assessments and adversarial robustness. Science-specialized models showed significant deficiencies in logical and ethical reasoning capabilities, along with concerning vulnerabilities in safety evaluations, particularly in high-risk domains such as biosecurity and chemical weapons. By open-sourcing our framework, we provide a foundation for developing more trustworthy AI systems and advancing research on model safety and ethics in scientific contexts.

arXiv:2510.25992v1 Announce Type: cross Abstract: Large Language Models (LLMs) often struggle with problems that require multi-step reasoning. For small-scale open-source models, Reinforcement Learning with Verifiable Rewards (RLVR) fails when correct solutions are rarely sampled even after many attempts, while Supervised Fine-Tuning (SFT) tends to overfit long demonstrations through rigid token-by-token imitation. To address this gap, we propose Supervised Reinforcement Learning (SRL), a framework that reformulates problem solving as generating a sequence of logical "actions". SRL trains the model to generate an internal reasoning monologue before committing to each action. It provides smoother rewards based on the similarity between the model's actions and expert actions extracted from the SFT dataset in a step-wise manner. This supervision offers richer learning signals even when all rollouts are incorrect, while encouraging flexible reasoning guided by expert demonstrations. As a result, SRL enables small models to learn challenging problems previously unlearnable by SFT or RLVR. Moreover, initializing training with SRL before refining with RLVR yields the strongest overall performance. Beyond reasoning benchmarks, SRL generalizes effectively to agentic software engineering tasks, establishing it as a robust and versatile training framework for reasoning-oriented LLMs.

arXiv:2510.26151v1 Announce Type: cross Abstract: Large annotated datasets are essential for training robust Computer-Aided Diagnosis (CAD) models for breast cancer detection or risk prediction. However, acquiring such datasets with fine-detailed annotation is both costly and time-consuming. Vision-Language Models (VLMs), such as CLIP, which are pre-trained on large image-text pairs, offer a promising solution by enhancing robustness and data efficiency in medical imaging tasks. This paper introduces a novel Multi-View Mammography and Language Model for breast cancer classification and risk prediction, trained on a dataset of paired mammogram images and synthetic radiology reports. Our MV-MLM leverages multi-view supervision to learn rich representations from extensive radiology data by employing cross-modal self-supervision across image-text pairs. This includes multiple views and the corresponding pseudo-radiology reports. We propose a novel joint visual-textual learning strategy to enhance generalization and accuracy performance over different data types and tasks to distinguish breast tissues or cancer characteristics(calcification, mass) and utilize these patterns to understand mammography images and predict cancer risk. We evaluated our method on both private and publicly available datasets, demonstrating that the proposed model achieves state-of-the-art performance in three classification tasks: (1) malignancy classification, (2) subtype classification, and (3) image-based cancer risk prediction. Furthermore, the model exhibits strong data efficiency, outperforming existing fully supervised or VLM baselines while trained on synthetic text reports and without the need for actual radiology reports.

arXiv:2410.14879v4 Announce Type: replace-cross Abstract: Passive tracking methods, such as phone and wearable sensing, have become dominant in monitoring human behaviors in modern ubiquitous computing studies. While there have been significant advances in machine-learning approaches to translate periods of raw sensor data to model momentary behaviors, (e.g., physical activity recognition), there still remains a significant gap in the translation of these sensing streams into meaningful, high-level, context-aware insights that are required for various applications (e.g., summarizing an individual's daily routine). To bridge this gap, experts often need to employ a context-driven sensemaking process in real-world studies to derive insights. This process often requires manual effort and can be challenging even for experienced researchers due to the complexity of human behaviors. We conducted three rounds of user studies with 21 experts to explore solutions to address challenges with sensemaking. We follow a human-centered design process to identify needs and design, iterate, build, and evaluate Vital Insight (VI), a novel, LLM-assisted, prototype system to enable human-in-the-loop inference (sensemaking) and visualizations of multi-modal passive sensing data from smartphones and wearables. Using the prototype as a technology probe, we observe experts' interactions with it and develop an expert sensemaking model that explains how experts move between direct data representations and AI-supported inferences to explore, question, and validate insights. Through this iterative process, we also synthesize and discuss a list of design implications for the design of future AI-augmented visualization systems to better assist experts' sensemaking processes in multi-modal health sensing data.

Nature, Published online: 29 October 2025; doi:10.1038/s41586-025-09686-5

This multi-omic longitudinal analysis of the healthy human peripheral immune system constructs the Human Immune Health Atlas and assembles data on immune cell composition and state changes with age, including responses to cytomegalovirus infection and influenza vaccination.

arXiv:2510.24528v1 Announce Type: new Abstract: The capability of in-context learning (ICL) enables large language models (LLMs) to perform novel tasks without parameter updates by conditioning on a few input-output examples. However, collecting high-quality examples for new or challenging tasks can be costly and labor-intensive. In this work, we propose a cost-efficient two-stage pipeline that reduces reliance on LLMs for data labeling. Our approach first leverages readily available cross-task examples to prompt an LLM and pseudo-label a small set of target task instances. We then introduce a graph-based label propagation method that spreads label information to the remaining target examples without additional LLM queries. The resulting fully pseudo-labeled dataset is used to construct in-task demonstrations for ICL. This pipeline combines the flexibility of cross-task supervision with the scalability of LLM-free propagation. Experiments across five tasks demonstrate that our method achieves strong performance while lowering labeling costs.

arXiv:2510.24701v1 Announce Type: cross Abstract: We present Tongyi DeepResearch, an agentic large language model, which is specifically designed for long-horizon, deep information-seeking research tasks. To incentivize autonomous deep research agency, Tongyi DeepResearch is developed through an end-to-end training framework that combines agentic mid-training and agentic post-training, enabling scalable reasoning and information seeking across complex tasks. We design a highly scalable data synthesis pipeline that is fully automatic, without relying on costly human annotation, and empowers all training stages. By constructing customized environments for each stage, our system enables stable and consistent interactions throughout. Tongyi DeepResearch, featuring 30.5 billion total parameters, with only 3.3 billion activated per token, achieves state-of-the-art performance across a range of agentic deep research benchmarks, including Humanity's Last Exam, BrowseComp, BrowseComp-ZH, WebWalkerQA, xbench-DeepSearch, FRAMES and xbench-DeepSearch-2510. We open-source the model, framework, and complete solutions to empower the community.

arXiv:2510.22003v1 Announce Type: cross Abstract: Generative AI (GenAI) is reshaping enterprise architecture work in agile software organizations, yet evidence on its effects remains scattered. We report a systematic literature review (SLR), following established SLR protocols of Kitchenham and PRISMA, of 1,697 records, yielding 33 studies across enterprise, solution, domain, business, and IT architect roles. GenAI most consistently supports (i) design ideation and trade-off exploration; (ii) rapid creation and refinement of artifacts (e.g., code, models, documentation); and (iii) architectural decision support and knowledge retrieval. Reported risks include opacity and bias, contextually incorrect outputs leading to rework, privacy and compliance concerns, and social loafing. We also identify emerging skills and competencies, including prompt engineering, model evaluation, and professional oversight, and organizational enablers around readiness and adaptive governance. The review contributes with (1) a mapping of GenAI use cases and risks in agile architecting, (2) implications for capability building and governance, and (3) an initial research agenda on human-AI collaboration in architecture. Overall, the findings inform responsible adoption of GenAI that accelerates digital transformation while safeguarding architectural integrity.

arXiv:2510.22242v1 Announce Type: cross Abstract: Large Language Models (LLMs) increasingly serve as research assistants, yet their reliability in scholarly tasks remains under-evaluated. In this work, we introduce PaperAsk, a benchmark that systematically evaluates LLMs across four key research tasks: citation retrieval, content extraction, paper discovery, and claim verification. We evaluate GPT-4o, GPT-5, and Gemini-2.5-Flash under realistic usage conditions-via web interfaces where search operations are opaque to the user. Through controlled experiments, we find consistent reliability failures: citation retrieval fails in 48-98% of multi-reference queries, section-specific content extraction fails in 72-91% of cases, and topical paper discovery yields F1 scores below 0.32, missing over 60% of relevant literature. Further human analysis attributes these failures to the uncontrolled expansion of retrieved context and the tendency of LLMs to prioritize semantically relevant text over task instructions. Across basic tasks, the LLMs display distinct failure behaviors: ChatGPT often withholds responses rather than risk errors, whereas Gemini produces fluent but fabricated answers. To address these issues, we develop lightweight reliability classifiers trained on PaperAsk data to identify unreliable outputs. PaperAsk provides a reproducible and diagnostic framework for advancing the reliability evaluation of LLM-based scholarly assistance systems.

arXiv:2510.22620v1 Announce Type: cross Abstract: AI agents powered by large language models (LLMs) are being deployed at scale, yet we lack a systematic understanding of how the choice of backbone LLM affects agent security. The non-deterministic sequential nature of AI agents complicates security modeling, while the integration of traditional software with AI components entangles novel LLM vulnerabilities with conventional security risks. Existing frameworks only partially address these challenges as they either capture specific vulnerabilities only or require modeling of complete agents. To address these limitations, we introduce threat snapshots: a framework that isolates specific states in an agent's execution flow where LLM vulnerabilities manifest, enabling the systematic identification and categorization of security risks that propagate from the LLM to the agent level. We apply this framework to construct the $\operatorname{b}^3$ benchmark, a security benchmark based on 194331 unique crowdsourced adversarial attacks. We then evaluate 31 popular LLMs with it, revealing, among other insights, that enhanced reasoning capabilities improve security, while model size does not correlate with security. We release our benchmark, dataset, and evaluation code to facilitate widespread adoption by LLM providers and practitioners, offering guidance for agent developers and incentivizing model developers to prioritize backbone security improvements.

arXiv:2510.23241v1 Announce Type: cross Abstract: In this work, we introduce Progressive Growing of Patch Size, an automatic curriculum learning approach for 3D medical image segmentation. Our approach progressively increases the patch size during model training, resulting in an improved class balance for smaller patch sizes and accelerated convergence of the training process. We evaluate our curriculum approach in two settings: a resource-efficient mode and a performance mode, both regarding Dice score performance and computational costs across 15 diverse and popular 3D medical image segmentation tasks. The resource-efficient mode matches the Dice score performance of the conventional constant patch size sampling baseline with a notable reduction in training time to only 44%. The performance mode improves upon constant patch size segmentation results, achieving a statistically significant relative mean performance gain of 1.28% in Dice Score. Remarkably, across all 15 tasks, our proposed performance mode manages to surpass the constant patch size baseline in Dice Score performance, while simultaneously reducing training time to only 89%. The benefits are particularly pronounced for highly imbalanced tasks such as lesion segmentation tasks. Rigorous experiments demonstrate that our performance mode not only improves mean segmentation performance but also reduces performance variance, yielding more trustworthy model comparison. Furthermore, our findings reveal that the proposed curriculum sampling is not tied to a specific architecture but represents a broadly applicable strategy that consistently boosts performance across diverse segmentation models, including UNet, UNETR, and SwinUNETR. In summary, we show that this simple yet elegant transformation on input data substantially improves both Dice Score performance and training runtime, while being compatible across diverse segmentation backbones.

arXiv:2505.17895v2 Announce Type: replace-cross Abstract: The quality of foundation models depends heavily on their training data. Consequently, great efforts have been put into dataset curation. Yet most approaches rely on manual tuning of coarse-grained mixtures of large buckets of data, or filtering by hand-crafted heuristics. An approach that is ultimately more scalable (let alone more satisfying) is to \emph{learn} which data is actually valuable for training. This type of meta-learning could allow more sophisticated, fine-grained, and effective curation. Our proposed \emph{DataRater} is an instance of this idea. It estimates the value of training on any particular data point. This is done by meta-learning using `meta-gradients', with the objective of improving training efficiency on held out data. In extensive experiments across a range of model scales and datasets, we find that using our DataRater to filter data is highly effective, resulting in significantly improved compute efficiency.

arXiv:2505.18087v2 Announce Type: replace-cross Abstract: Recent progress in Large Vision-Language Models (LVLMs) has enabled promising applications in medical tasks, such as report generation and visual question answering. However, existing benchmarks focus mainly on the final diagnostic answer, offering limited insight into whether models engage in clinically meaningful reasoning. To address this, we present CheXStruct and CXReasonBench, a structured pipeline and benchmark built on the publicly available MIMIC-CXR-JPG dataset. CheXStruct automatically derives a sequence of intermediate reasoning steps directly from chest X-rays, such as segmenting anatomical regions, deriving anatomical landmarks and diagnostic measurements, computing diagnostic indices, and applying clinical thresholds. CXReasonBench leverages this pipeline to evaluate whether models can perform clinically valid reasoning steps and to what extent they can learn from structured guidance, enabling fine-grained and transparent assessment of diagnostic reasoning. The benchmark comprises 18,988 QA pairs across 12 diagnostic tasks and 1,200 cases, each paired with up to 4 visual inputs, and supports multi-path, multi-stage evaluation including visual grounding via anatomical region selection and diagnostic measurements. Even the strongest of 12 evaluated LVLMs struggle with structured reasoning and generalization, often failing to link abstract knowledge with anatomically grounded visual interpretation. The code is available at https://github.com/ttumyche/CXReasonBench

arXiv:2506.06522v2 Announce Type: replace-cross Abstract: Recent work on large language models (LLMs) has increasingly focused on post-training and alignment with datasets curated to enhance instruction following, world knowledge, and specialized skills. However, most post-training datasets used in leading open- and closed-source LLMs remain inaccessible to the public, with limited information about their construction process. This lack of transparency has motivated the recent development of open-source post-training corpora. While training on these open alternatives can yield performance comparable to that of leading models, systematic comparisons remain challenging due to the significant computational cost of conducting them rigorously at scale, and are therefore largely absent. As a result, it remains unclear how specific samples, task types, or curation strategies influence downstream performance when assessing data quality. In this work, we conduct the first comprehensive side-by-side analysis of two prominent open post-training datasets: Tulu-3-SFT-Mix and SmolTalk. Using the Magpie framework, we annotate each sample with detailed quality metrics, including turn structure (single-turn vs. multi-turn), task category, input quality, and response quality, and we derive statistics that reveal structural and qualitative similarities and differences between the two datasets. Based on these insights, we design a principled curation recipe that produces a new data mixture, TuluTalk, which contains 14% fewer samples than either source dataset while matching or exceeding their performance on key benchmarks. Our findings offer actionable insights for constructing more effective post-training datasets that improve model performance within practical resource limits. To support future research, we publicly release both the annotated source datasets and our curated TuluTalk mixture.

arXiv:2507.05177v3 Announce Type: replace-cross Abstract: Empathetic interaction is a cornerstone of human-machine communication, due to the need for understanding speech enriched with paralinguistic cues and generating emotional and expressive responses. However, the most powerful empathetic LSLMs are increasingly closed off, leaving the crucial details about the architecture, data and development opaque to researchers. Given the critical need for transparent research into the LSLMs and empathetic behavior, we present OpenS2S, a fully open-source, transparent and end-to-end LSLM designed to enable empathetic speech interactions. Based on our empathetic speech-to-text model BLSP-Emo, OpenS2S further employs a streaming interleaved decoding architecture to achieve low-latency speech generation. To facilitate end-to-end training, OpenS2S incorporates an automated data construction pipeline that synthesizes diverse, high-quality empathetic speech dialogues at low cost. By leveraging large language models to generate empathetic content and controllable text-to-speech systems to introduce speaker and emotional variation, we construct a scalable training corpus with rich paralinguistic diversity and minimal human supervision. We release the fully open-source OpenS2S model, including the dataset, model weights, pre-training and fine-tuning codes, to empower the broader research community and accelerate innovation in empathetic speech systems. The project webpage can be accessed at https://casia-lm.github.io/OpenS2S

npj Digital Medicine, Published online: 27 October 2025; doi:10.1038/s41746-025-02093-0

The rise of digital health technologies has provided individuals with unprecedented access to biometric data and health insights. However, excess monitoring may contribute to fatigue, anxiety, and information overload, sometimes reducing engagement and worsening outcomes. This article explores how artificial intelligence-enabled assistants might help address this challenge by filtering, contextualizing, and personalizing health information, potentially supporting informed self-management while mitigating some unintended harms of digital health technologies.

arXiv:2510.21285v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) demonstrate remarkable capabilities on complex reasoning tasks but remain vulnerable to severe safety risks, including harmful content generation and jailbreak attacks. Existing mitigation strategies rely on injecting heuristic safety signals during training, which often suppress reasoning ability and fail to resolve the safety-reasoning trade-off. To systematically investigate this issue, we analyze the reasoning trajectories of diverse LRMs and uncover a phenomenon we term Self-Jailbreak, where models override their own risk assessments and justify responding to unsafe prompts. This finding reveals that LRMs inherently possess the ability to reject unsafe queries, but this ability is compromised, resulting in harmful outputs. Building on these insights, we propose the Chain-of-Guardrail (CoG), a training framework that recomposes or backtracks unsafe reasoning steps, steering the model back onto safe trajectories while preserving valid reasoning chains. Extensive experiments across multiple reasoning and safety benchmarks demonstrate that CoG substantially improves the safety of current LRMs while preserving comparable reasoning ability, significantly outperforming prior methods that suffer from severe safety-reasoning trade-offs.

arXiv:2510.21652v1 Announce Type: new Abstract: AI agents hold the potential to revolutionize scientific productivity by automating literature reviews, replicating experiments, analyzing data, and even proposing new directions of inquiry; indeed, there are now many such agents, ranging from general-purpose "deep research" systems to specialized science-specific agents, such as AI Scientist and AIGS. Rigorous evaluation of these agents is critical for progress. Yet existing benchmarks fall short on several fronts: they (1) fail to provide holistic, product-informed measures of real-world use cases such as science research; (2) lack reproducible agent tools necessary for a controlled comparison of core agentic capabilities; (3) do not account for confounding variables such as model cost and tool access; (4) do not provide standardized interfaces for quick agent prototyping and evaluation; and (5) lack comprehensive baseline agents necessary to identify true advances. In response, we define principles and tooling for more rigorously benchmarking agents. Using these, we present AstaBench, a suite that provides the first holistic measure of agentic ability to perform scientific research, comprising 2400+ problems spanning the entire scientific discovery process and multiple scientific domains, and including many problems inspired by actual user requests to deployed Asta agents. Our suite comes with the first scientific research environment with production-grade search tools that enable controlled, reproducible evaluation, better accounting for confounders. Alongside, we provide a comprehensive suite of nine science-optimized classes of Asta agents and numerous baselines. Our extensive evaluation of 57 agents across 22 agent classes reveals several interesting findings, most importantly that despite meaningful progress on certain individual aspects, AI remains far from solving the challenge of science research assistance.

arXiv:2510.20099v1 Announce Type: new Abstract: We present AI PB, a production-scale generative agent deployed in real retail finance. Unlike reactive chatbots that answer queries passively, AI PB proactively generates grounded, compliant, and user-specific investment insights. It integrates (i) a component-based orchestration layer that deterministically routes between internal and external LLMs based on data sensitivity, (ii) a hybrid retrieval pipeline using OpenSearch and the finance-domain embedding model, and (iii) a multi-stage recommendation mechanism combining rule heuristics, sequential behavioral modeling, and contextual bandits. Operating fully on-premises under Korean financial regulations, the system employs Docker Swarm and vLLM across 24 X NVIDIA H100 GPUs. Through human QA and system metrics, we demonstrate that grounded generation with explicit routing and layered safety can deliver trustworthy AI insights in high-stakes finance.