arXiv:2508.15126v2 Announce Type: replace Abstract: Recent advances in large language models (LLMs) have enabled AI agents to autonomously generate scientific proposals, conduct experiments, author papers, and perform peer reviews. Yet this flood of AI-generated research content collides with a fragmented and largely closed publication ecosystem. Traditional journals and conferences rely on human peer review, making them difficult to scale and often reluctant to accept AI-generated research content; existing preprint servers (e.g. arXiv) lack rigorous quality-control mechanisms. Consequently, a significant amount of high-quality AI-generated research lacks appropriate venues for dissemination, hindering its potential to advance scientific progress. To address these challenges, we introduce aiXiv, a next-generation open-access platform for human and AI scientists. Its multi-agent architecture allows research proposals and papers to be submitted, reviewed, and iteratively refined by both human and AI scientists. It also provides API and MCP interfaces that enable seamless integration of heterogeneous human and AI scientists, creating a scalable and extensible ecosystem for autonomous scientific discovery. Through extensive experiments, we demonstrate that aiXiv is a reliable and robust platform that significantly enhances the quality of AI-generated research proposals and papers after iterative revising and reviewing on aiXiv. Our work lays the groundwork for a next-generation open-access ecosystem for AI scientists, accelerating the publication and dissemination of high-quality AI-generated research content. Code: https://github.com/aixiv-org aiXiv: https://aixiv.science

arXiv:2512.13747v1 Announce Type: cross Abstract: With the rapid progress of large language models (LLMs), advanced multimodal large language models (MLLMs) have demonstrated impressive zero-shot capabilities on vision-language tasks. In the biomedical domain, however, even state-of-the-art MLLMs struggle with basic Medical Decision Making (MDM) tasks. We investigate this limitation using two challenging datasets: (1) three-stage Alzheimer's disease (AD) classification (normal, mild cognitive impairment, dementia), where category differences are visually subtle, and (2) MIMIC-CXR chest radiograph classification with 14 non-mutually exclusive conditions. Our empirical study shows that text-only reasoning consistently outperforms vision-only or vision-text settings, with multimodal inputs often performing worse than text alone. To mitigate this, we explore three strategies: (1) in-context learning with reason-annotated exemplars, (2) vision captioning followed by text-only inference, and (3) few-shot fine-tuning of the vision tower with classification supervision. These findings reveal that current MLLMs lack grounded visual understanding and point to promising directions for improving multimodal decision making in healthcare.

Background: Precision health promotion, which aims to tailor health messages to individual needs, is hampered by the lack of structured metadata in vast digital health resource libraries. This bottleneck prevents scalable, personalized content delivery and exacerbates information overload for the public. Objective: This study aimed to develop, deploy, and validate an automated tagging system using a large language model (LLM) to create the foundational metadata infrastructure required for tailored health communication at scale. Methods: We developed a comprehensive, 3-tier health promotion taxonomy (10 primary, 34 secondary, and 90,562 tertiary tags) using a hybrid Delphi and corpus-mining methodology. We then constructed a hybrid inference pipeline by fine-tuning a Baichuan2-7B LLM with low-rank adaptation for initial tag generation. This was then refined by a domain-specific named entity recognition model and standardized against a vector database. The system’s performance was evaluated against manual annotations from nonexpert staff on a test set of 1000 resources. We used a “no gold standard” framework, comparing the artificial intelligence–human (A-H) interrater reliability (IRR) with a supplemental human-human (H-H) IRR baseline and expert adjudication for cases where artificial intelligence provided additional tags (“AI Additive”). Results: The A-H agreement was moderate (Cohen κ=0.54, 95% CI 0.53-0.56; Jaccard similarity coefficient=0.48, 95% CI 0.46-0.50). Critically, this was higher than the baseline nonexpert H-H agreement (Cohen κ=0.32, 95% CI 0.29-0.35; Jaccard similarity coefficient=0.35, 95% CI 0.27-0.43). A granular analysis of disagreements revealed that in 15.9% (159/1000) of the cases, the “AI Additive” tags were not identified by human annotators. Expert adjudication of these cases confirmed that the “AI Additive” tags were correct and relevant with a precision of 90% (45/50; 95% CI 78.2%-96.7%). Conclusions: A fine-tuned LLM, integrated into a hybrid pipeline, can function as a powerful augmentation tool for health content annotation. The system’s consistency (A-H κ=0.54) was found to be superior to the baseline human workflow (H-H κ=0.32). By moving beyond simple automation to reliably identify relevant health topics missed by manual annotators with high, expert-validated accuracy, this study provides a robust technical and methodological blueprint for implementing artificial intelligence to enhance precision health communication in public health settings.