Yiwei Li

Zhengliang Liu, Yue Huang, Xiaowei Yu et al.

The digitization of healthcare has facilitated the sharing and re-using of medical data but has also raised concerns about confidentiality and privacy. HIPAA (Health Insurance Portability and Accountability Act) mandates removing re-identifying information before the dissemination of medical records. Thus, effective and efficient solutions for de-identifying medical data, especially those in free-text forms, are highly needed. While various computer-assisted de-identification methods, including both rule-based and learning-based, have been developed and used in prior practice, such solutions still lack generalizability or need to be fine-tuned according to different scenarios, significantly imposing restrictions in wider use. The advancement of large language models (LLM), such as ChatGPT and GPT-4, have shown great potential in processing text data in the medical domain with zero-shot in-context learning, especially in the task of privacy protection, as these models can identify confidential information by their powerful named entity recognition (NER) capability. In this work, we developed a novel GPT4-enabled de-identification framework (``DeID-GPT") to automatically identify and remove the identifying information. Compared to existing commonly used medical text data de-identification methods, our developed DeID-GPT showed the highest accuracy and remarkable reliability in masking private information from the unstructured medical text while preserving the original structure and meaning of the text. This study is one of the earliest to utilize ChatGPT and GPT-4 for medical text data processing and de-identification, which provides insights for further research and solution development on the use of LLMs such as ChatGPT/GPT-4 in healthcare. Codes and benchmarking data information are available at https://github.com/yhydhx/ChatGPT-API.

Zhengliang Liu, Aoxiao Zhong, Yiwei Li et al.

We introduce Radiology-GPT, a large language model for radiology. Using an instruction tuning approach on an extensive dataset of radiology domain knowledge, Radiology-GPT demonstrates superior performance compared to general language models such as StableLM, Dolly and LLaMA. It exhibits significant versatility in radiological diagnosis, research, and communication. This work serves as a catalyst for future developments in clinical NLP. The successful implementation of Radiology-GPT is indicative of the potential of localizing generative large language models, specifically tailored for distinctive medical specialties, while ensuring adherence to privacy standards such as HIPAA. The prospect of developing individualized, large-scale language models that cater to specific needs of various hospitals presents a promising direction. The fusion of conversational competence and domain-specific knowledge in these models is set to foster future development in healthcare AI. A demo of Radiology-GPT is available at https://huggingface.co/spaces/allen-eric/radiology-gpt.

Haixing Dai, Chong Ma, Zhiling Yan et al.

This paper introduces SAMAug, a novel visual point augmentation method for the Segment Anything Model (SAM) that enhances interactive image segmentation performance. SAMAug generates augmented point prompts to provide more information about the user's intention to SAM. Starting with an initial point prompt, SAM produces an initial mask, which is then fed into our proposed SAMAug to generate augmented point prompts. By incorporating these extra points, SAM can generate augmented segmentation masks based on both the augmented point prompts and the initial prompt, resulting in improved segmentation performance. We conducted evaluations using four different point augmentation strategies: random sampling, sampling based on maximum difference entropy, maximum distance, and saliency. Experiment results on the COCO, Fundus, COVID QUEx, and ISIC2018 datasets show that SAMAug can boost SAM's segmentation results, especially using the maximum distance and saliency. SAMAug demonstrates the potential of visual prompt augmentation for computer vision. Codes of SAMAug are available at github.com/yhydhx/SAMAug

Haixing Dai, Yiwei Li, Zhengliang Liu et al.

In this pioneering study, inspired by AutoGPT, the state-of-the-art open-source application based on the GPT-4 large language model, we develop a novel tool called AD-AutoGPT which can conduct data collection, processing, and analysis about complex health narratives of Alzheimer's Disease in an autonomous manner via users' textual prompts. We collated comprehensive data from a variety of news sources, including the Alzheimer's Association, BBC, Mayo Clinic, and the National Institute on Aging since June 2022, leading to the autonomous execution of robust trend analyses, intertopic distance maps visualization, and identification of salient terms pertinent to Alzheimer's Disease. This approach has yielded not only a quantifiable metric of relevant discourse but also valuable insights into public focus on Alzheimer's Disease. This application of AD-AutoGPT in public health signifies the transformative potential of AI in facilitating a data-rich understanding of complex health narratives like Alzheimer's Disease in an autonomous manner, setting the groundwork for future AI-driven investigations in global health landscapes.

Kimi Team, Tongtong Bai, Yifan Bai et al.

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework that dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently. Extensive evaluations show that Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks. Agent Swarm also reduces latency by up to $4.5\times$ over single-agent baselines. We release the post-trained Kimi K2.5 model checkpoint to facilitate future research and real-world applications of agentic intelligence.

Zhengliang Liu, Yiwei Li, Peng Shu et al.

This paper introduces Radiology-Llama2, a large language model specialized for radiology through a process known as instruction tuning. Radiology-Llama2 is based on the Llama2 architecture and further trained on a large dataset of radiology reports to generate coherent and clinically useful impressions from radiological findings. Quantitative evaluations using ROUGE metrics on the MIMIC-CXR and OpenI datasets demonstrate that Radiology-Llama2 achieves state-of-the-art performance compared to other generative language models, with a Rouge-1 score of 0.4834 on MIMIC-CXR and 0.4185 on OpenI. Additional assessments by radiology experts highlight the model's strengths in understandability, coherence, relevance, conciseness, and clinical utility. The work illustrates the potential of localized language models designed and tuned for specialized domains like radiology. When properly evaluated and deployed, such models can transform fields like radiology by automating rote tasks and enhancing human expertise.

Xiang Li, Lin Zhao, Lu Zhang et al.

Large-scale Artificial General Intelligence (AGI) models, including Large Language Models (LLMs) such as ChatGPT/GPT-4, have achieved unprecedented success in a variety of general domain tasks. Yet, when applied directly to specialized domains like medical imaging, which require in-depth expertise, these models face notable challenges arising from the medical field's inherent complexities and unique characteristics. In this review, we delve into the potential applications of AGI models in medical imaging and healthcare, with a primary focus on LLMs, Large Vision Models, and Large Multimodal Models. We provide a thorough overview of the key features and enabling techniques of LLMs and AGI, and further examine the roadmaps guiding the evolution and implementation of AGI models in the medical sector, summarizing their present applications, potentialities, and associated challenges. In addition, we highlight potential future research directions, offering a holistic view on upcoming ventures. This comprehensive review aims to offer insights into the future implications of AGI in medical imaging, healthcare, and beyond.

Rong Zhou, Zhengqing Yuan, Zhiling Yan et al.

Biomedical image segmentation is crucial for accurately diagnosing and analyzing various diseases. However, Convolutional Neural Networks (CNNs) and Transformers, the most commonly used architectures for this task, struggle to effectively capture long-range dependencies due to the inherent locality of CNNs and the computational complexity of Transformers. To address this limitation, we introduce TTT-Unet, a novel framework that integrates Test-Time Training (TTT) layers into the traditional U-Net architecture for biomedical image segmentation. TTT-Unet dynamically adjusts model parameters during the testing time, enhancing the model's ability to capture both local and long-range features. We evaluate TTT-Unet on multiple medical imaging datasets, including 3D abdominal organ segmentation in CT and MR images, instrument segmentation in endoscopy images, and cell segmentation in microscopy images. The results demonstrate that TTT-Unet consistently outperforms state-of-the-art CNN-based and Transformer-based segmentation models across all tasks. The code is available at https://github.com/rongzhou7/TTT-Unet.

Zhengliang Liu, Peilong Wang, Yiwei Li et al.

This paper presents RadOnc-GPT, a large language model specialized for radiation oncology through advanced tuning methods. RadOnc-GPT was finetuned on a large dataset of radiation oncology patient records from the Mayo Clinic in Arizona. The model employs instruction tuning on three key tasks - generating radiotherapy treatment regimens, determining optimal radiation modalities, and providing diagnostic descriptions/ICD codes based on patient diagnostic details. Evaluations conducted by comparing RadOnc-GPT outputs to general large language model outputs showed higher ROUGE scores in these three tasks. The study demonstrated the potential of using large language models fine-tuned using domain-specific knowledge like RadOnc-GPT to achieve transformational capabilities in highly specialized healthcare fields such as radiation oncology. However, our model's clinical relevance requires confirmation, and it specializes in only the aforementioned three specific tasks and lacks broader applicability. Furthermore, its evaluation through ROUGE scores might not reflect the true semantic and clinical accuracy - challenges we intend to address in future research.

Chenhao Tang, Zhengliang Liu, Chong Ma et al.

Privacy policies serve as the primary conduit through which online service providers inform users about their data collection and usage procedures. However, in a bid to be comprehensive and mitigate legal risks, these policy documents are often quite verbose. In practical use, users tend to click the Agree button directly rather than reading them carefully. This practice exposes users to risks of privacy leakage and legal issues. Recently, the advent of Large Language Models (LLM) such as ChatGPT and GPT-4 has opened new possibilities for text analysis, especially for lengthy documents like privacy policies. In this study, we investigate a privacy policy text analysis framework PolicyGPT based on the LLM. This framework was tested using two datasets. The first dataset comprises of privacy policies from 115 websites, which were meticulously annotated by legal experts, categorizing each segment into one of 10 classes. The second dataset consists of privacy policies from 304 popular mobile applications, with each sentence manually annotated and classified into one of another 10 categories. Under zero-shot learning conditions, PolicyGPT demonstrated robust performance. For the first dataset, it achieved an accuracy rate of 97%, while for the second dataset, it attained an 87% accuracy rate, surpassing that of the baseline machine learning and neural network models.

Zhengliang Liu, Tianyang Zhong, Yiwei Li et al.

The rise of large language models (LLMs) has marked a pivotal shift in the field of natural language processing (NLP). LLMs have revolutionized a multitude of domains, and they have made a significant impact in the medical field. Large language models are now more abundant than ever, and many of these models exhibit bilingual capabilities, proficient in both English and Chinese. However, a comprehensive evaluation of these models remains to be conducted. This lack of assessment is especially apparent within the context of radiology NLP. This study seeks to bridge this gap by critically evaluating thirty two LLMs in interpreting radiology reports, a crucial component of radiology NLP. Specifically, the ability to derive impressions from radiologic findings is assessed. The outcomes of this evaluation provide key insights into the performance, strengths, and weaknesses of these LLMs, informing their practical applications within the medical domain.

Tianyang Zhong, Zhengliang Liu, Yi Pan et al.

This comprehensive study evaluates the performance of OpenAI's o1-preview large language model across a diverse array of complex reasoning tasks, spanning multiple domains, including computer science, mathematics, natural sciences, medicine, linguistics, and social sciences. Through rigorous testing, o1-preview demonstrated remarkable capabilities, often achieving human-level or superior performance in areas ranging from coding challenges to scientific reasoning and from language processing to creative problem-solving. Key findings include: -83.3% success rate in solving complex competitive programming problems, surpassing many human experts. -Superior ability in generating coherent and accurate radiology reports, outperforming other evaluated models. -100% accuracy in high school-level mathematical reasoning tasks, providing detailed step-by-step solutions. -Advanced natural language inference capabilities across general and specialized domains like medicine. -Impressive performance in chip design tasks, outperforming specialized models in areas such as EDA script generation and bug analysis. -Remarkable proficiency in anthropology and geology, demonstrating deep understanding and reasoning in these specialized fields. -Strong capabilities in quantitative investing. O1 has comprehensive financial knowledge and statistical modeling skills. -Effective performance in social media analysis, including sentiment analysis and emotion recognition. The model excelled particularly in tasks requiring intricate reasoning and knowledge integration across various fields. While some limitations were observed, including occasional errors on simpler problems and challenges with certain highly specialized concepts, the overall results indicate significant progress towards artificial general intelligence.

Siheng Li, Yichun Yin, Cheng Yang et al. · pku

Hot news is one of the most popular topics in daily conversations. However, news grounded conversation has long been stymied by the lack of well-designed task definition and scarce data. In this paper, we propose a novel task, Proactive News Grounded Conversation, in which a dialogue system can proactively lead the conversation based on some key topics of the news. In addition, both information-seeking and chit-chat scenarios are included realistically, where the user may ask a series of questions about the news details or express their opinions and be eager to chat. To further develop this novel task, we collect a human-to-human Chinese dialogue dataset \ts{NewsDialogues}, which includes 1K conversations with a total of 14.6K utterances and detailed annotations for target topics and knowledge spans. Furthermore, we propose a method named Predict-Generate-Rank, consisting of a generator for grounded knowledge prediction and response generation, and a ranker for the ranking of multiple responses to alleviate the exposure bias. We conduct comprehensive experiments to demonstrate the effectiveness of the proposed method and further present several key findings and challenges to prompt future research.

Zhengliang Liu, Weihang You, Peng Shu et al.

American college applications require students to navigate fragmented admissions policies, repetitive and conditional forms, and ambiguous questions that often demand cross-referencing multiple sources. We present EZCollegeApp, a large language model (LLM)-powered system that assists high-school students by structuring application forms, grounding suggested answers in authoritative admissions documents, and maintaining full human control over final responses. The system introduces a mapping-first paradigm that separates form understanding from answer generation, enabling consistent reasoning across heterogeneous application portals. EZCollegeApp integrates document ingestion from official admissions websites, retrieval-augmented question answering, and a human-in-the-loop chatbot interface that presents suggestions alongside application fields without automated submission. We describe the system architecture, data pipeline, internal representations, security and privacy measures, and evaluation through automated testing and human quality assessment. Our source code is released on GitHub (https://github.com/ezcollegeapp-public/ezcollegeapp-public) to facilitate the broader impact of this work.

Tianyang Zhong, Wei Zhao, Yutong Zhang et al.

Radiology report generation, as a key step in medical image analysis, is critical to the quantitative analysis of clinically informed decision-making levels. However, complex and diverse radiology reports with cross-source heterogeneity pose a huge generalizability challenge to the current methods under massive data volume, mainly because the style and normativity of radiology reports are obviously distinctive among institutions, body regions inspected and radiologists. Recently, the advent of large language models (LLM) offers great potential for recognizing signs of health conditions. To resolve the above problem, we collaborate with the Second Xiangya Hospital in China and propose ChatRadio-Valuer based on the LLM, a tailored model for automatic radiology report generation that learns generalizable representations and provides a basis pattern for model adaptation in sophisticated analysts' cases. Specifically, ChatRadio-Valuer is trained based on the radiology reports from a single institution by means of supervised fine-tuning, and then adapted to disease diagnosis tasks for human multi-system evaluation (i.e., chest, abdomen, muscle-skeleton, head, and maxillofacial $\&$ neck) from six different institutions in clinical-level events. The clinical dataset utilized in this study encompasses a remarkable total of \textbf{332,673} observations. From the comprehensive results on engineering indicators, clinical efficacy and deployment cost metrics, it can be shown that ChatRadio-Valuer consistently outperforms state-of-the-art models, especially ChatGPT (GPT-3.5-Turbo) and GPT-4 et al., in terms of the diseases diagnosis from radiology reports. ChatRadio-Valuer provides an effective avenue to boost model generalization performance and alleviate the annotation workload of experts to enable the promotion of clinical AI applications in radiology reports.

Zhengliang Liu, Hanqi Jiang, Tianyang Zhong et al.

In this paper, we present a large-scale evaluation probing GPT-4V's capabilities and limitations for biomedical image analysis. GPT-4V represents a breakthrough in artificial general intelligence (AGI) for computer vision, with applications in the biomedical domain. We assess GPT-4V's performance across 16 medical imaging categories, including radiology, oncology, ophthalmology, pathology, and more. Tasks include modality recognition, anatomy localization, disease diagnosis, report generation, and lesion detection. The extensive experiments provide insights into GPT-4V's strengths and weaknesses. Results show GPT-4V's proficiency in modality and anatomy recognition but difficulty with disease diagnosis and localization. GPT-4V excels at diagnostic report generation, indicating strong image captioning skills. While promising for biomedical imaging AI, GPT-4V requires further enhancement and validation before clinical deployment. We emphasize responsible development and testing for trustworthy integration of biomedical AGI. This rigorous evaluation of GPT-4V on diverse medical images advances understanding of multimodal large language models (LLMs) and guides future work toward impactful healthcare applications.

Yiwei Li, Shuai Wang, Chong-Yung Chi et al.

In this paper, we investigate federated clustering (FedC) problem, that aims to accurately partition unlabeled data samples distributed over massive clients into finite clusters under the orchestration of a parameter server, meanwhile considering data privacy. Though it is an NP-hard optimization problem involving real variables denoting cluster centroids and binary variables denoting the cluster membership of each data sample, we judiciously reformulate the FedC problem into a non-convex optimization problem with only one convex constraint, accordingly yielding a soft clustering solution. Then a novel FedC algorithm using differential privacy (DP) technique, referred to as DP-FedC, is proposed in which partial clients participation and multiple local model updating steps are also considered. Furthermore, various attributes of the proposed DP-FedC are obtained through theoretical analyses of privacy protection and convergence rate, especially for the case of non-identically and independently distributed (non-i.i.d.) data, that ideally serve as the guidelines for the design of the proposed DP-FedC. Then some experimental results on two real datasets are provided to demonstrate the efficacy of the proposed DP-FedC together with its much superior performance over some state-of-the-art FedC algorithms, and the consistency with all the presented analytical results.

Huawen Hu, Enze Shi, Chenxi Yue et al.

Human-in-the-loop reinforcement learning integrates human expertise to accelerate agent learning and provide critical guidance and feedback in complex fields. However, many existing approaches focus on single-agent tasks and require continuous human involvement during the training process, significantly increasing the human workload and limiting scalability. In this paper, we propose HARP (Human-Assisted Regrouping with Permutation Invariant Critic), a multi-agent reinforcement learning framework designed for group-oriented tasks. HARP integrates automatic agent regrouping with strategic human assistance during deployment, enabling and allowing non-experts to offer effective guidance with minimal intervention. During training, agents dynamically adjust their groupings to optimize collaborative task completion. When deployed, they actively seek human assistance and utilize the Permutation Invariant Group Critic to evaluate and refine human-proposed groupings, allowing non-expert users to contribute valuable suggestions. In multiple collaboration scenarios, our approach is able to leverage limited guidance from non-experts and enhance performance. The project can be found at https://github.com/huawen-hu/HARP.

Zhiling Yan, Weixiang Sun, Rong Zhou et al.

Medical image segmentation and video object segmentation are essential for diagnosing and analyzing diseases by identifying and measuring biological structures. Recent advances in natural domain have been driven by foundation models like the Segment Anything Model 2 (SAM-2). To explore the performance of SAM-2 in biomedical applications, we designed three evaluation pipelines for single-frame 2D image segmentation, multi-frame 3D image segmentation and multi-frame video segmentation with varied prompt designs, revealing SAM-2's limitations in medical contexts. Consequently, we developed BioSAM-2, an enhanced foundation model optimized for biomedical data based on SAM-2. Our experiments show that BioSAM-2 not only surpasses the performance of existing state-of-the-art foundation models but also matches or even exceeds specialist models, demonstrating its efficacy and potential in the medical domain.

Yiwei Li, Shuai Wang, Tsung-Hui Chang et al.

Federated learning (FL) is a new paradigm that enables many clients to jointly train a machine learning (ML) model under the orchestration of a parameter server while keeping the local data not being exposed to any third party. However, the training of FL is an interactive process between local clients and the parameter server. Such process would cause privacy leakage since adversaries may retrieve sensitive information by analyzing the overheard messages. In this paper, we propose a new federated stochastic primal-dual algorithm with differential privacy (FedSPD-DP). Compared to the existing methods, the proposed FedSPD-DP incorporates local stochastic gradient descent (local SGD) and partial client participation (PCP) for addressing the issues of communication efficiency and straggler effects due to randomly accessed clients. Our analysis shows that the data sampling strategy and PCP can enhance the data privacy whereas the larger number of local SGD steps could increase privacy leakage, revealing a non-trivial tradeoff between algorithm communication efficiency and privacy protection. Specifically, we show that, by guaranteeing $(ε, δ)$-DP for each client per communication round, the proposed algorithm guarantees $(\mathcal{O}(qε\sqrt{p T}), δ)$-DP after $T$ communication rounds while maintaining an $\mathcal{O}(1/\sqrt{pTQ})$ convergence rate for a convex and non-smooth learning problem, where $Q$ is the number of local SGD steps, $p$ is the client sampling probability, $q=\max_{i} q_i/\sqrt{1-q_i}$ and $q_i$ is the data sampling probability of each client under PCP. Experiment results are presented to evaluate the practical performance of the proposed algorithm and comparison with state-of-the-art methods.

Zhengliang Liu, Yiwei Li, Qian Cao et al.

Recent advances in artificial general intelligence (AGI), particularly large language models and creative image generation systems have demonstrated impressive capabilities on diverse tasks spanning the arts and humanities. However, the swift evolution of AGI has also raised critical questions about its responsible deployment in these culturally significant domains traditionally seen as profoundly human. This paper provides a comprehensive analysis of the applications and implications of AGI for text, graphics, audio, and video pertaining to arts and the humanities. We survey cutting-edge systems and their usage in areas ranging from poetry to history, marketing to film, and communication to classical art. We outline substantial concerns pertaining to factuality, toxicity, biases, and public safety in AGI systems, and propose mitigation strategies. The paper argues for multi-stakeholder collaboration to ensure AGI promotes creativity, knowledge, and cultural values without undermining truth or human dignity. Our timely contribution summarizes a rapidly developing field, highlighting promising directions while advocating for responsible progress centering on human flourishing. The analysis lays the groundwork for further research on aligning AGI's technological capacities with enduring social goods.

Yanjun Lyu, Zihao Wu, Lu Zhang et al.

Pre-trained large language models(LLMs) have attracted increasing attention in biomedical domains due to their success in natural language processing. However, the complex traits and heterogeneity of multi-sources genomics data pose significant challenges when adapting these models to the bioinformatics and biomedical field. To address these challenges, we present GP-GPT, the first specialized large language model for genetic-phenotype knowledge representation and genomics relation analysis. Our model is fine-tuned in two stages on a comprehensive corpus composed of over 3,000,000 terms in genomics, proteomics, and medical genetics, derived from multiple large-scale validated datasets and scientific publications. GP-GPT demonstrates proficiency in accurately retrieving medical genetics information and performing common genomics analysis tasks, such as genomics information retrieval and relationship determination. Comparative experiments across domain-specific tasks reveal that GP-GPT outperforms state-of-the-art LLMs, including Llama2, Llama3 and GPT-4. These results highlight GP-GPT's potential to enhance genetic disease relation research and facilitate accurate and efficient analysis in the fields of genomics and medical genetics. Our investigation demonstrated the subtle changes of bio-factor entities' representations in the GP-GPT, which suggested the opportunities for the application of LLMs to advancing gene-phenotype research.

Yiwei Li, Chien-Wei Huang, Shuai Wang et al.

Federated learning (FL) has been recognized as a rapidly growing research area, where the model is trained over massively distributed clients under the orchestration of a parameter server (PS) without sharing clients' data. This paper delves into a class of federated problems characterized by non-convex and non-smooth loss functions, that are prevalent in FL applications but challenging to handle due to their intricate non-convexity and non-smoothness nature and the conflicting requirements on communication efficiency and privacy protection. In this paper, we propose a novel federated primal-dual algorithm with bidirectional model sparsification tailored for non-convex and non-smooth FL problems, and differential privacy is applied for privacy guarantee. Its unique insightful properties and some privacy and convergence analyses are also presented as the FL algorithm design guidelines. Extensive experiments on real-world data are conducted to demonstrate the effectiveness of the proposed algorithm and much superior performance than some state-of-the-art FL algorithms, together with the validation of all the analytical results and properties.

Xinyu Gong, Jason Holmes, Yiwei Li et al.

Recent advances in Large Language Models (LLMs) have presented new opportunities for integrating Artificial General Intelligence (AGI) into biological research and education. This study evaluated the capabilities of leading LLMs, including GPT-4, GPT-3.5, PaLM2, Claude2, and SenseNova, in answering conceptual biology questions. The models were tested on a 108-question multiple-choice exam covering biology topics in molecular biology, biological techniques, metabolic engineering, and synthetic biology. Among the models, GPT-4 achieved the highest average score of 90 and demonstrated the greatest consistency across trials with different prompts. The results indicated GPT-4's proficiency in logical reasoning and its potential to aid biology research through capabilities like data analysis, hypothesis generation, and knowledge integration. However, further development and validation are still required before the promise of LLMs in accelerating biological discovery can be realized.

Jason Holmes, Rui Peng, Yiwei Li et al.

IMPORTANCE The response effectiveness of different large language models (LLMs) and various individuals, including medical students, graduate students, and practicing physicians, in pediatric ophthalmology consultations, has not been clearly established yet. OBJECTIVE Design a 100-question exam based on pediatric ophthalmology to evaluate the performance of LLMs in highly specialized scenarios and compare them with the performance of medical students and physicians at different levels. DESIGN, SETTING, AND PARTICIPANTS This survey study assessed three LLMs, namely ChatGPT (GPT-3.5), GPT-4, and PaLM2, were assessed alongside three human cohorts: medical students, postgraduate students, and attending physicians, in their ability to answer questions related to pediatric ophthalmology. It was conducted by administering questionnaires in the form of test papers through the LLM network interface, with the valuable participation of volunteers. MAIN OUTCOMES AND MEASURES Mean scores of LLM and humans on 100 multiple-choice questions, as well as the answer stability, correlation, and response confidence of each LLM. RESULTS GPT-4 performed comparably to attending physicians, while ChatGPT (GPT-3.5) and PaLM2 outperformed medical students but slightly trailed behind postgraduate students. Furthermore, GPT-4 exhibited greater stability and confidence when responding to inquiries compared to ChatGPT (GPT-3.5) and PaLM2. CONCLUSIONS AND RELEVANCE Our results underscore the potential for LLMs to provide medical assistance in pediatric ophthalmology and suggest significant capacity to guide the education of medical students.

Kimi Team, Yu Zhang, Zongyu Lin et al.

We introduce Kimi Linear, a hybrid linear attention architecture that, for the first time, outperforms full attention under fair comparisons across various scenarios -- including short-context, long-context, and reinforcement learning (RL) scaling regimes. At its core lies Kimi Delta Attention (KDA), an expressive linear attention module that extends Gated DeltaNet with a finer-grained gating mechanism, enabling more effective use of limited finite-state RNN memory. Our bespoke chunkwise algorithm achieves high hardware efficiency through a specialized variant of the Diagonal-Plus-Low-Rank (DPLR) transition matrices, which substantially reduces computation compared to the general DPLR formulation while remaining more consistent with the classical delta rule. We pretrain a Kimi Linear model with 3B activated parameters and 48B total parameters, based on a layerwise hybrid of KDA and Multi-Head Latent Attention (MLA). Our experiments show that with an identical training recipe, Kimi Linear outperforms full MLA with a sizeable margin across all evaluated tasks, while reducing KV cache usage by up to 75% and achieving up to 6 times decoding throughput for a 1M context. These results demonstrate that Kimi Linear can be a drop-in replacement for full attention architectures with superior performance and efficiency, including tasks with longer input and output lengths. To support further research, we open-source the KDA kernel and vLLM implementations, and release the pre-trained and instruction-tuned model checkpoints.

Zhengliang Liu, Yiwei Li, Oleksandra Zolotarevych et al.

Large language models have demonstrated remarkable capabilities in natural language processing, yet their application to political discourse analysis remains underexplored. This paper introduces a novel approach to evaluating presidential debate performances using LLMs, addressing the longstanding challenge of objectively assessing debate outcomes. We propose a framework that analyzes candidates' "Policies, Persona, and Perspective" (3P) and how they resonate with the "Interests, Ideologies, and Identity" (3I) of four key audience groups: voters, businesses, donors, and politicians. Our method employs large language models to generate the LLM-POTUS Score, a quantitative measure of debate performance based on the alignment between 3P and 3I. We apply this framework to analyze transcripts from recent U.S. presidential debates, demonstrating its ability to provide nuanced, multi-dimensional assessments of candidate performances. Our results reveal insights into the effectiveness of different debating strategies and their impact on various audience segments. This study not only offers a new tool for political analysis but also explores the potential and limitations of using LLMs as impartial judges in complex social contexts. In addition, this framework provides individual citizens with an independent tool to evaluate presidential debate performances, which enhances democratic engagement and reduces reliance on potentially biased media interpretations and institutional influence, thereby strengthening the foundation of informed civic participation.

Jason Holmes, Shuyuan Ye, Yiwei Li et al.

Purpose: The performance of three different large language models (LLMS) (GPT-3.5, GPT-4, and PaLM2) in answering ophthalmology professional questions was evaluated and compared with that of three different professional populations (medical undergraduates, medical masters, and attending physicians). Methods: A 100-item ophthalmology single-choice test was administered to three different LLMs (GPT-3.5, GPT-4, and PaLM2) and three different professional levels (medical undergraduates, medical masters, and attending physicians), respectively. The performance of LLM was comprehensively evaluated and compared with the human group in terms of average score, stability, and confidence. Results: Each LLM outperformed undergraduates in general, with GPT-3.5 and PaLM2 being slightly below the master's level, while GPT-4 showed a level comparable to that of attending physicians. In addition, GPT-4 showed significantly higher answer stability and confidence than GPT-3.5 and PaLM2. Conclusion: Our study shows that LLM represented by GPT-4 performs better in the field of ophthalmology. With further improvements, LLM will bring unexpected benefits in medical education and clinical decision making in the near future.

Zhenyuan Yang, Xuhui Lin, Qinyi He et al.

The emergence of Large Language Models (LLMs) and multimodal foundation models (FMs) has generated heightened interest in their applications that integrate vision and language. This paper investigates the capabilities of ChatGPT-4V and Gemini Pro for Street View Imagery, Built Environment, and Interior by evaluating their performance across various tasks. The assessments include street furniture identification, pedestrian and car counts, and road width measurement in Street View Imagery; building function classification, building age analysis, building height analysis, and building structure classification in the Built Environment; and interior room classification, interior design style analysis, interior furniture counts, and interior length measurement in Interior. The results reveal proficiency in length measurement, style analysis, question answering, and basic image understanding, but highlight limitations in detailed recognition and counting tasks. While zero-shot learning shows potential, performance varies depending on the problem domains and image complexities. This study provides new insights into the strengths and weaknesses of multimodal foundation models for practical challenges in Street View Imagery, Built Environment, and Interior. Overall, the findings demonstrate foundational multimodal intelligence, emphasizing the potential of FMs to drive forward interdisciplinary applications at the intersection of computer vision and language.

Xinglin Wang, Shaoxiong Feng, Yiwei Li et al.

Self-consistency (SC), a widely used decoding strategy for chain-of-thought reasoning, shows significant gains across various multi-step reasoning tasks but comes with a high cost due to multiple sampling with the preset size. Its variants, Adaptive self-consistency (ASC) and Early-stopping self-consistency (ESC), dynamically adjust the number of samples based on the posterior distribution of a set of pre-samples, reducing the cost of SC with minimal impact on performance. Both methods, however, do not exploit the prior information about question difficulty. It often results in unnecessary repeated sampling for easy questions that could be accurately answered with just one attempt, wasting resources. To tackle this problem, we propose Difficulty-Adaptive Self-Consistency (DSC), which leverages the difficulty information of batch queries from both prior and posterior perspectives to adaptively allocate inference resources, further reducing the overall cost of SC. To demonstrate the effectiveness of DSC, we conduct extensive experiments on three popular categories of reasoning tasks: arithmetic, commonsense and symbolic reasoning on six benchmarks. The empirical results show that DSC consistently surpasses the strong baseline ASC and ESC in terms of costs by a significant margin, while attaining comparable performances.

Sohan Shankar, Yi Pan, Hanqi Jiang et al.

This position and survey paper identifies the emerging convergence of neuroscience, artificial general intelligence (AGI), and neuromorphic computing toward a unified research paradigm. Using a framework grounded in brain physiology, we highlight how synaptic plasticity, sparse spike-based communication, and multimodal association provide design principles for next-generation AGI systems that potentially combine both human and machine intelligences. The review traces this evolution from early connectionist models to state-of-the-art large language models, demonstrating how key innovations like transformer attention, foundation-model pre-training, and multi-agent architectures mirror neurobiological processes like cortical mechanisms, working memory, and episodic consolidation. We then discuss emerging physical substrates capable of breaking the von Neumann bottleneck to achieve brain-scale efficiency in silicon: memristive crossbars, in-memory compute arrays, and emerging quantum and photonic devices. There are four critical challenges at this intersection: 1) integrating spiking dynamics with foundation models, 2) maintaining lifelong plasticity without catastrophic forgetting, 3) unifying language with sensorimotor learning in embodied agents, and 4) enforcing ethical safeguards in advanced neuromorphic autonomous systems. This combined perspective across neuroscience, computation, and hardware offers an integrative agenda for in each of these fields.

Shougao Zhang, Mengqi Zhou, Yuxi Wang et al.

Urban areas, as the primary human habitat in modern civilization, accommodate a broad spectrum of social activities. With the surge of embodied intelligence, recent years have witnessed an increasing presence of physical agents in urban areas, such as autonomous vehicles and delivery robots. As a result, practitioners significantly value crafting authentic, simulation-ready 3D cities to facilitate the training and verification of such agents. However, this task is quite challenging. Current generative methods fall short in either diversity, controllability, or fidelity. In this work, we resort to the procedural content generation (PCG) technique for high-fidelity generation. It assembles superior assets according to empirical rules, ultimately leading to industrial-grade outcomes. To ensure diverse and self contained creation, we design a management protocol to accommodate extensive PCG plugins with distinct functions and interfaces. Based on this unified PCG library, we develop a multi-agent framework to transform multi-modal instructions, including OSM, semantic maps, and satellite images, into executable programs. The programs coordinate relevant plugins to construct the 3D city consistent with the control condition. A visual feedback scheme is introduced to further refine the initial outcomes. Our method, named CityX, demonstrates its superiority in creating diverse, controllable, and realistic 3D urban scenes. The synthetic scenes can be seamlessly deployed as a real-time simulator and an infinite data generator for embodied intelligence research. Our project page: https://cityx-lab.github.io.

Xinglin Wang, Hao Lin, Shaoxiong Feng et al.

Test-Time Scaling (TTS) enhances the reasoning capabilities of large language models by allocating additional inference compute to explore the solution space. However, existing parallel TTS methods typically keep branches isolated during search: intermediate discoveries remain branch-private and cannot guide other branches in time. This information isolation causes substantial redundant exploration, as branches repeatedly rediscover information already found elsewhere and require more search steps to collect complete decision information needed to reach correct answers. To bridge this gap, we propose \textbf{Collaborative Parallel Thinking (CPT)}, a training-free inference framework that enables search-time information sharing across parallel branches. CPT extracts compact intermediate information from ongoing branches, maintains a deduplicated query-level information pool, and broadcasts pool entries through the input context, allowing each branch in subsequent search steps to reuse discoveries made by other branches rather than rediscover the same information. Empirically, experiments on HMMT and AIME benchmarks show that CPT establishes a stronger accuracy--latency Pareto frontier than strong baselines across rollout budgets and model scales, highlighting search-time collaboration as an effective direction for efficient parallel TTS.

Bin Sun, Yitong Li, Fei Mi et al.

Conditional variational models, using either continuous or discrete latent variables, are powerful for open-domain dialogue response generation. However, previous works show that continuous latent variables tend to reduce the coherence of generated responses. In this paper, we also found that discrete latent variables have difficulty capturing more diverse expressions. To tackle these problems, we combine the merits of both continuous and discrete latent variables and propose a Hybrid Latent Variable (HLV) method. Specifically, HLV constrains the global semantics of responses through discrete latent variables and enriches responses with continuous latent variables. Thus, we diversify the generated responses while maintaining relevance and coherence. In addition, we propose Conditional Hybrid Variational Transformer (CHVT) to construct and to utilize HLV with transformers for dialogue generation. Through fine-grained symbolic-level semantic information and additive Gaussian mixing, we construct the distribution of continuous variables, prompting the generation of diverse expressions. Meanwhile, to maintain the relevance and coherence, the discrete latent variable is optimized by self-separation training. Experimental results on two dialogue generation datasets (DailyDialog and Opensubtitles) show that CHVT is superior to traditional transformer-based variational mechanism w.r.t. diversity, relevance and coherence metrics. Moreover, we also demonstrate the benefit of applying HLV to fine-tuning two pre-trained dialogue models (PLATO and BART-base).

Yue Huang, Chujie Gao, Siyuan Wu et al.

Generative Foundation Models (GenFMs) have emerged as transformative tools. However, their widespread adoption raises critical concerns regarding trustworthiness across dimensions. This paper presents a comprehensive framework to address these challenges through three key contributions. First, we systematically review global AI governance laws and policies from governments and regulatory bodies, as well as industry practices and standards. Based on this analysis, we propose a set of guiding principles for GenFMs, developed through extensive multidisciplinary collaboration that integrates technical, ethical, legal, and societal perspectives. Second, we introduce TrustGen, the first dynamic benchmarking platform designed to evaluate trustworthiness across multiple dimensions and model types, including text-to-image, large language, and vision-language models. TrustGen leverages modular components--metadata curation, test case generation, and contextual variation--to enable adaptive and iterative assessments, overcoming the limitations of static evaluation methods. Using TrustGen, we reveal significant progress in trustworthiness while identifying persistent challenges. Finally, we provide an in-depth discussion of the challenges and future directions for trustworthy GenFMs, which reveals the complex, evolving nature of trustworthiness, highlighting the nuanced trade-offs between utility and trustworthiness, and consideration for various downstream applications, identifying persistent challenges and providing a strategic roadmap for future research. This work establishes a holistic framework for advancing trustworthiness in GenAI, paving the way for safer and more responsible integration of GenFMs into critical applications. To facilitate advancement in the community, we release the toolkit for dynamic evaluation.

Xinglin Wang, Yiwei Li, Shaoxiong Feng et al.

Self-consistency (SC), leveraging multiple samples from LLMs, shows significant gains on various reasoning tasks but struggles with free-form generation due to the difficulty of aggregating answers. Its variants, UCS and USC, rely on sample selection or voting mechanisms to improve output quality. These methods, however, face limitations due to their inability to fully utilize the nuanced consensus knowledge present within multiple candidate samples, often resulting in suboptimal outputs. We propose Fine-Grained Self-Consistency (FSC) to addresses these limitations by extracting and integrating segment-level commonalities from candidate samples, enhancing the performance of LLMs both in open-ended and reasoning tasks. Based on this, we present two additional strategies: candidate filtering, which enhances overall quality by identifying highly similar candidate sets, and merging, which reduces input token requirements by combining similar samples. The effectiveness of FSC is demonstrated through extensive experiments on various tasks, including summarization, code generation, and mathematical reasoning, using GPT-3.5-turbo and GPT-4. The results indicate significant improvements over baseline methods, showcasing the potential of FSC to optimize output quality by effectively synthesizing fine-grained consensus knowledge from multiple samples.

Yiwei Li, Shaoxiong Feng, Bin Sun et al.

Generative dialogue models suffer badly from the generic response problem, limiting their applications to a few toy scenarios. Recently, an interesting approach, namely negative training, has been proposed to alleviate this problem by reminding the model not to generate high-frequency responses during training. However, its performance is hindered by two issues, ignoring low-frequency but generic responses and bringing low-frequency but meaningless responses. In this paper, we propose a novel negative training paradigm, called negative distillation, to keep the model away from the undesirable generic responses while avoiding the above problems. First, we introduce a negative teacher model that can produce query-wise generic responses, and then the student model is required to maximize the distance with multi-level negative knowledge. Empirical results show that our method outperforms previous negative training methods significantly.

Bin Sun, Shaoxiong Feng, Yiwei Li et al.

Complex dialogue mappings (CDM), including one-to-many and many-to-one mappings, tend to make dialogue models generate incoherent or dull responses, and modeling these mappings remains a huge challenge for neural dialogue systems. To alleviate these problems, methods like introducing external information, reconstructing the optimization function, and manipulating data samples are proposed, while they primarily focus on avoiding training with CDM, inevitably weakening the model's ability of understanding CDM in human conversations and limiting further improvements in model performance. This paper proposes a Sentence Semantic \textbf{Seg}mentation guided \textbf{C}onditional \textbf{V}ariational \textbf{A}uto-\textbf{E}ncoder (SegCVAE) method which can model and take advantages of the CDM data. Specifically, to tackle the incoherent problem caused by one-to-many, SegCVAE uses response-related prominent semantics to constrained the latent variable. To mitigate the non-diverse problem brought by many-to-one, SegCVAE segments multiple prominent semantics to enrich the latent variables. Three novel components, Internal Separation, External Guidance, and Semantic Norms, are proposed to achieve SegCVAE. On dialogue generation tasks, both the automatic and human evaluation results show that SegCVAE achieves new state-of-the-art performance.

Peiwen Yuan, Shaoxiong Feng, Yiwei Li et al.

In-Context Learning (ICL) enables large language models (LLMs) to achieve rapid task adaptation by learning from demonstrations. With the increase in available context length of LLMs, recent experiments have shown that the performance of ICL does not necessarily scale well in many-shot (demonstration) settings. We theoretically and experimentally confirm that the reason lies in more demonstrations dispersing the model attention from the query, hindering its understanding of key content. Inspired by how humans learn from examples, we propose a training-free method FocusICL, which conducts triviality filtering to avoid attention being diverted by unimportant contents at token-level and operates hierarchical attention to further ensure sufficient attention towards current query at demonstration-level. We also design an efficient hyperparameter searching strategy for FocusICL based on model perplexity of demonstrations. Comprehensive experiments validate that FocusICL achieves an average performance improvement of 5.2% over vanilla ICL and scales well with many-shot demonstrations.

Kimi Team, Yifan Bai, Yiping Bao et al. · tsinghua

We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.

Yiwei Li, Bin Sun, Shaoxiong Feng et al.

There is a growing interest in improving the conversational ability of models by filtering the raw dialogue corpora. Previous filtering strategies usually rely on a scoring method to assess and discard samples from one perspective, enabling the model to enhance the corresponding dialogue attributes (e.g., consistency) more easily. However, the discarded samples may obtain high scores in other perspectives and can provide regularization effects on the model learning, which causes the performance improvement to be sensitive to the filtering ratio. In this work, we propose a multi-view attribute-enhanced dialogue learning framework that strengthens the attribute-related features more robustly and comprehensively. Instead of filtering the raw dataset to train the model, our framework first pre-trains the model on the raw dataset and then fine-tunes it through adapters on the selected sub-sets, which also enhances certain attributes of responses but without suffering from the problems mentioned above. Considering the variety of the dialogue attribute, we further design a multi-view enhancement mechanism, including multi-view selection and inter-view fusion. It groups the high-quality samples from multiple perspectives, respectively, and enhances different attributes of responses with the corresponding sample sets and adapters, keeping knowledge independent and allowing flexible integration. Empirical results and analysis show that our framework can improve the performance significantly in terms of enhancing dialogue attributes and fusing view-specific knowledge.

Yiwei Li, Jiayi Shi, Shaoxiong Feng et al.

Instruction data is crucial for improving the capability of Large Language Models (LLMs) to align with human-level performance. Recent research LIMA demonstrates that alignment is essentially a process where the model adapts instructions' interaction style or format to solve various tasks, leveraging pre-trained knowledge and skills. Therefore, for instructional data, the most important aspect is the task it represents, rather than the specific semantics and knowledge information. The latent representations of instructions play roles for some instruction-related tasks like data selection and demonstrations retrieval. However, they are always derived from text embeddings, encompass overall semantic information that influences the representation of task categories. In this work, we introduce a new concept, instruction embedding, and construct Instruction Embedding Benchmark (IEB) for its training and evaluation. Then, we propose a baseline Prompt-based Instruction Embedding (PIE) method to make the representations more attention on tasks. The evaluation of PIE, alongside other embedding methods on IEB with two designed tasks, demonstrates its superior performance in accurately identifying task categories. Moreover, the application of instruction embeddings in four downstream tasks showcases its effectiveness and suitability for instruction-related tasks.

Yiwei Li, Shaoxiong Feng, Bin Sun et al.

With the development of deep learning, advanced dialogue generation methods usually require a greater amount of computational resources. One promising approach to obtaining a high-performance and lightweight model is knowledge distillation, which relies heavily on the pre-trained powerful teacher. Collaborative learning, also known as online knowledge distillation, is an effective way to conduct one-stage group distillation in the absence of a well-trained large teacher model. However, previous work has a severe branch homogeneity problem due to the same training objective and the independent identical training sets. To alleviate this problem, we consider the dialogue attributes in the training of network branches. Each branch learns the attribute-related features based on the selected subset. Furthermore, we propose a dual group-based knowledge distillation method, consisting of positive distillation and negative distillation, to further diversify the features of different branches in a steadily and interpretable way. The proposed approach significantly improves branch heterogeneity and outperforms state-of-the-art collaborative learning methods on two widely used open-domain dialogue datasets.

Xinglin Wang, Peiwen Yuan, Shaoxiong Feng et al.

Piaget's Theory of Cognitive Development (PTC) posits that the development of cognitive levels forms the foundation for human learning across various abilities. As Large Language Models (LLMs) have recently shown remarkable abilities across a wide variety of tasks, we are curious about the cognitive levels of current LLMs: to what extent they have developed and how this development has been achieved. To this end, we construct a benchmark CogLM (Cognitive Ability Evaluation for Language Model) based on PTC to assess the cognitive levels of LLMs. CogLM comprises 1,220 questions spanning 10 cognitive abilities crafted by more than 20 human experts, providing a comprehensive testbed for the cognitive levels of LLMs. Through extensive experiments across multiple mainstream LLMs with CogLM, we find that: (1) In our testing framework, advanced LLMs (such as GPT-4) have demonstrated human-like cognitive abilities, comparable to those of a 20-year-old human. (2) The parameter size and optimization objective are two key factors affecting the cognitive levels of LLMs. (3) The performance on downstream tasks is positively correlated with the level of cognitive abilities. These findings fill the gap in research on the cognitive abilities of LLMs, tracing the development of LLMs from a cognitive perspective and guiding the future direction of their evolution.

Peiwen Yuan, Shaoxiong Feng, Yiwei Li et al.

The guidance from capability evaluations has greatly propelled the progress of both human society and Artificial Intelligence. However, as LLMs evolve, it becomes challenging to construct evaluation benchmarks for them with accurate labels on hard tasks that approach the boundaries of human capabilities. To credibly conduct evaluation without accurate labels (denoted as poor-supervised evaluation), we propose the PoEM framework. We first prove that the capability of a model can be equivalently assessed by the consistency between it and certain reference model, when their prediction distributions are independent and the sample size is infinite. To alleviate the insufficiencies of the conditions in reality, we further introduce an algorithm that treats humans (when available) and the models under evaluation as reference models, alternately conducting model weights calibration and filtering during E-step and M-step. Comprehensive experiments across 3 types of tasks with 16 mainstream LLMs have shown that PoEM under poor supervision can achieve an average of 0.98 Pearson correlation coefficient with supervised evaluation results, demonstrating good effectiveness, efficiency and generalizability. More generally, PoEM has advanced the evaluation paradigm evolution from human-centric to human&model-centric by treating both of them as reference models, mitigating the limitations of human evaluation in the era of LLMs.

Zihao Wu, Steven Xu, Bowen Chen et al.

With the emergence of large language models (LLMs) and AI agent frameworks, the human-AI co-work paradigm known as Vibe Coding is changing how people code, making it more accessible and productive. In scientific research, where workflows are more complex and the burden of specialized labor limits independent researchers and those in low-resource areas, the potential impact is even greater, particularly in biomedicine, which involves heterogeneous data modalities and multi-step analytical pipelines. In this paper, we introduce Vibe Medicine, a co-work paradigm in which clinicians and researchers direct skill-augmented AI agents through natural language to execute complex, multi-step biomedical workflows, while retaining the role of research director who specifies objectives, reviews intermediate results, and makes domain-informed decisions. The enabling infrastructure consists of three layers: capable LLMs, agent frameworks such as OpenClaw and Hermes Agent, and the OpenClaw medical skills collection, which includes more than 1,000 curated skills from multiple open-source repositories. We analyze the architecture and skill categories of this collection across ten biomedical domains, and present case studies covering rare disease diagnosis, drug repurposing, and clinical trial design that demonstrate end-to-end workflows in practice. We also identify the principal risks, such as hallucination, data privacy, and over-reliance, and outline directions toward more reliable, trustworthy, and clinically integrated agent-assisted research that advances research and technological equity and reduces health care resource disparities.

Xinglin Wang, Jiayi Shi, Shaoxiong Feng et al.

Test-Time Scaling enhances the reasoning capabilities of Large Language Models by allocating additional inference compute to broaden the exploration of the solution space. However, existing search strategies typically treat rollouts as disposable samples, where valuable intermediate insights are effectively discarded after each trial. This systemic memorylessness leads to massive computational redundancy, as models repeatedly re-derive discovered conclusions and revisit known dead ends across extensive attempts. To bridge this gap, we propose \textbf{Recycling Search Experience (RSE)}, a self-guided, training-free strategy that turns test-time search from a series of isolated trials into a cumulative process. By actively distilling raw trajectories into a shared experience bank, RSE enables positive recycling of intermediate conclusions to shortcut redundant derivations and negative recycling of failure patterns to prune encountered dead ends. Theoretically, we provide an analysis that formalizes the efficiency gains of RSE, validating its advantage over independent sampling in solving complex reasoning tasks. Empirically, extensive experiments on HMMT24, HMMT25, IMO-Bench, and HLE show that RSE consistently outperforms strong baselines with comparable computational cost, achieving state-of-the-art scaling efficiency.

Hang Yin, Xiaomin He, PeiWen Yuan et al.

Existing vision-language models often suffer from spatial hallucinations, i.e., generating incorrect descriptions about the relative positions of objects in an image. We argue that this problem mainly stems from the asymmetric properties between images and text. To enrich the spatial understanding ability of vision-language models, we propose a simple, annotation-free, plug-and-play method named $\text{Stitch and Tell}$ (abbreviated as SiTe), which injects structured spatial supervision into data. It constructs stitched image-text pairs by stitching images along a spatial axis and generating spatially-aware captions or question answer pairs based on the layout of stitched image, without relying on costly advanced models or human involvement. We evaluate SiTe across three architectures including LLaVA-v1.5-7B, LLaVA-Qwen2-1.5B and HALVA-7B, two training datasets, and eight benchmarks. Experiments show that SiTe improves spatial understanding tasks such as $\text{MME}_{\text{Position}}$ (+5.50%) and Spatial-MM (+4.19%), while maintaining or improving performance on general vision-language benchmarks including COCO-QA (+1.02%) and MMBench (+4.76%). Our findings suggest that explicitly injecting spatially-aware structure into training data offers an effective way to mitigate spatial hallucinations and improve spatial understanding, while preserving general vision-language capabilities.

Yiwei Li, Zihao Wu, Huaqin Zhao et al.

Radiologist eye-tracking data provide a rich record of how experts search, compare, and accumulate evidence during image reading; yet, existing methods exploit this signal only partially, either as a static spatial prior or as an auxiliary prediction target decoupled from diagnosis. We propose GazeWorld, a medical imaging world model that treats the image as the world and the radiologist's fixation sequence as a trajectory through it. GazeWorld autoregressively predicts the latent representation of the next fixated patch from all previously visited ones, while a spatial-completion branch covers unvisited regions. At inference, GazeWorld generates a sequence of patch representations from the image alone without requiring real gaze data. Frozen GazeWorld features achieve state-of-the-art diagnostic accuracy across all nine supervised settings on CheXpert, RSNA Pneumonia, and SIIM-ACR Pneumothorax, as well as the highest zero-shot accuracy on all three benchmarks. On the GazeSearch benchmark, a generic decoder trained on the same frozen features outperforms the purpose-built LogitGaze-Med by over 16\% in ScanMatch and 22\% in SED, despite not being explicitly trained to predict gaze. GazeWorld demonstrates that modeling how experts read, not just what they conclude, offers a promising pretraining paradigm for medical imaging AI.

Zhiling Yan, Sifan Song, Dingjie Song et al.

Recent "segment anything" efforts show promise by learning from large-scale data, but adapting such models directly to medical images remains challenging due to the complexity of medical data, noisy annotations, and continual learning requirements across diverse modalities and anatomical structures. In this work, we propose SAMed-2, a new foundation model for medical image segmentation built upon the SAM-2 architecture. Specifically, we introduce a temporal adapter into the image encoder to capture image correlations and a confidence-driven memory mechanism to store high-certainty features for later retrieval. This memory-based strategy counters the pervasive noise in large-scale medical datasets and mitigates catastrophic forgetting when encountering new tasks or modalities. To train and evaluate SAMed-2, we curate MedBank-100k, a comprehensive dataset spanning seven imaging modalities and 21 medical segmentation tasks. Our experiments on both internal benchmarks and 10 external datasets demonstrate superior performance over state-of-the-art baselines in multi-task scenarios. The code is available at: https://github.com/ZhilingYan/Medical-SAM-Bench.