Generalist Medical AI (GMAI) systems have demonstrated expert-level performance in biomedical perception tasks, yet their clinical utility remains limited by inadequate multi-modal explainability and suboptimal prognostic capabilities. Here, we present XMedGPT, a clinician-centric, multi-modal AI assistant that integrates textual and visual interpretability to support transparent and trustworthy medical decision-making. XMedGPT not only produces accurate diagnostic and descriptive outputs, but also grounds referenced anatomical sites within medical images, bridging critical gaps in interpretability and enhancing clinician usability. To support real-world deployment, we introduce a reliability indexing mechanism that quantifies uncertainty through consistency-based assessment via interactive question-answering. We validate XMedGPT across four pillars: multi-modal interpretability, uncertainty quantification, and prognostic modeling, and rigorous benchmarking. The model achieves an IoU of 0.703 across 141 anatomical regions, and a Kendall's tau-b of 0.479, demonstrating strong alignment between visual rationales and clinical outcomes. For uncertainty estimation, it attains an AUC of 0.862 on visual question answering and 0.764 on radiology report generation. In survival and recurrence prediction for lung and glioma cancers, it surpasses prior leading models by 26.9%, and outperforms GPT-4o by 25.0%. Rigorous benchmarking across 347 datasets covers 40 imaging modalities and external validation spans 4 anatomical systems confirming exceptional generalizability, with performance gains surpassing existing GMAI by 20.7% for in-domain evaluation and 16.7% on 11,530 in-house data evaluation. Together, XMedGPT represents a significant leap forward in clinician-centric AI integration, offering trustworthy and scalable support for diverse healthcare applications.

We present gg-bench, a collection of game environments designed to evaluate general reasoning capabilities in language models. Unlike most static benchmarks, gg-bench is a data generating process where new evaluation instances can be generated at will. In particular, gg-bench is synthetically generated by (1) using a large language model (LLM) to generate natural language descriptions of novel games, (2) using the LLM to implement each game in code as a Gym environment, and (3) training reinforcement learning (RL) agents via self-play on the generated games. We evaluate language models by their winrate against these RL agents by prompting models with the game description, current board state, and a list of valid moves, after which models output the moves they wish to take. gg-bench is challenging: state-of-the-art LLMs such as GPT-4o and Claude 3.7 Sonnet achieve winrates of 7-9% on gg-bench using in-context learning, while reasoning models such as o1, o3-mini and DeepSeek-R1 achieve average winrates of 31-36%. We release the generated games, data generation process, and evaluation code in order to support future modeling work and expansion of our benchmark.

Large language models (LLMs) have been widely used for relevance assessment in information retrieval. However, our study demonstrates that combining two distinct small language models (SLMs) with different architectures can outperform LLMs in this task. Our approach -- QUPID -- integrates a generative SLM with an embedding-based SLM, achieving higher relevance judgment accuracy while reducing computational costs compared to state-of-the-art LLM solutions. This computational efficiency makes QUPID highly scalable for real-world search systems processing millions of queries daily. In experiments across diverse document types, our method demonstrated consistent performance improvements (Cohen's Kappa of 0.646 versus 0.387 for leading LLMs) while offering 60x faster inference times. Furthermore, when integrated into production search pipelines, QUPID improved nDCG@5 scores by 1.9%. These findings underscore how architectural diversity in model combinations can significantly enhance both search relevance and operational efficiency in information retrieval systems.

Traditional topic models often struggle with contextual nuances and fail to adequately handle polysemy and rare words. This limitation typically results in topics that lack coherence and quality. Large Language Models (LLMs) can mitigate this issue by generating an initial set of topics. However, these raw topics frequently lack refinement and representativeness, which leads to redundancy without lexical similarity and reduced interpretability. This paper introduces HAMLET, a graph-driven architecture for cross-lingual healthcare topic modeling that uses LLMs. The proposed approach leverages neural-enhanced semantic fusion to refine the embeddings of topics generated by the LLM. Instead of relying solely on statistical co-occurrence or human interpretation to extract topics from a document corpus, this method introduces a topic embedding refinement that uses Bidirectional Encoder Representations from Transformers (BERT) and Graph Neural Networks (GNN). After topic generation, a hybrid technique that involves BERT and Sentence-BERT (SBERT) is employed for embedding. The topic representations are further refined using a GNN, which establishes connections between documents, topics, words, similar topics, and similar words. A novel method is introduced to compute similarities. Consequently, the topic embeddings are refined, and the top k topics are extracted. Experiments were conducted using two healthcare datasets, one in English and one in French, from which six sets were derived. The results demonstrate the effectiveness of HAMLET.

Large language models (LLMs) often generate fluent but factually incorrect outputs, known as hallucinations, which undermine their reliability in real-world applications. While uncertainty estimation has emerged as a promising strategy for detecting such errors, current metrics offer limited interpretability and lack clarity about the types of uncertainty they capture. In this paper, we present a systematic framework for decomposing LLM uncertainty into four distinct sources, inspired by previous research. We develop a source-specific estimation pipeline to quantify these uncertainty types and evaluate how existing metrics relate to each source across tasks and models. Our results show that metrics, task, and model exhibit systematic variation in uncertainty characteristic. Building on this, we propose a method for task specific metric/model selection guided by the alignment or divergence between their uncertainty characteristics and that of a given task. Our experiments across datasets and models demonstrate that our uncertainty-aware selection strategy consistently outperforms baseline strategies, helping us select appropriate models or uncertainty metrics, and contributing to more reliable and efficient deployment in uncertainty estimation.

With an increase in the capabilities of generative language models, a growing interest in embodied AI has followed. This contribution introduces RAI - a framework for creating embodied Multi Agent Systems for robotics. The proposed framework implements tools for Agents' integration with robotic stacks, Large Language Models, and simulations. It provides out-of-the-box integration with state-of-the-art systems like ROS 2. It also comes with dedicated mechanisms for the embodiment of Agents. These mechanisms have been tested on a physical robot, Husarion ROSBot XL, which was coupled with its digital twin, for rapid prototyping. Furthermore, these mechanisms have been deployed in two simulations: (1) robot arm manipulator and (2) tractor controller. All of these deployments have been evaluated in terms of their control capabilities, effectiveness of embodiment, and perception ability. The proposed framework has been used successfully to build systems with multiple agents. It has demonstrated effectiveness in all the aforementioned tasks. It also enabled identifying and addressing the shortcomings of the generative models used for embodied AI.

Text watermarking aims to subtly embed statistical signals into text by controlling the Large Language Model (LLM)'s sampling process, enabling watermark detectors to verify that the output was generated by the specified model. The robustness of these watermarking algorithms has become a key factor in evaluating their effectiveness. Current text watermarking algorithms embed watermarks in high-entropy tokens to ensure text quality. In this paper, we reveal that this seemingly benign design can be exploited by attackers, posing a significant risk to the robustness of the watermark. We introduce a generic efficient paraphrasing attack, the Self-Information Rewrite Attack (SIRA), which leverages the vulnerability by calculating the self-information of each token to identify potential pattern tokens and perform targeted attack. Our work exposes a widely prevalent vulnerability in current watermarking algorithms. The experimental results show SIRA achieves nearly 100% attack success rates on seven recent watermarking methods with only 0.88 USD per million tokens cost. Our approach does not require any access to the watermark algorithms or the watermarked LLM and can seamlessly transfer to any LLM as the attack model, even mobile-level models. Our findings highlight the urgent need for more robust watermarking.

With the widespread adoption of Mixture-of-Experts (MoE) models, there is a growing demand for efficient inference on memory-constrained devices. While offloading expert parameters to CPU memory and loading activated experts on demand has emerged as a potential solution, the large size of activated experts overburdens the limited PCIe bandwidth, hindering the effectiveness in latency-sensitive scenarios. To mitigate this, we propose FloE, an on-the-fly MoE inference system on memory-constrained GPUs. FloE is built on the insight that there exists substantial untapped redundancy within sparsely activated experts. It employs various compression techniques on the expert's internal parameter matrices to reduce the data movement load, combined with low-cost sparse prediction, achieving perceptible inference acceleration in wall-clock time on resource-constrained devices. Empirically, FloE achieves a 9.3x compression of parameters per expert in Mixtral-8x7B; enables deployment on a GPU with only 11GB VRAM, reducing the memory footprint by up to 8.5x; and delivers a 48.7x inference speedup compared to DeepSpeed-MII on a single GeForce RTX 3090 - all with only a 4.4$\%$ - 7.6$\%$ average performance degradation.

We introduce xGen-small, a family of 4B and 9B Transformer decoder models optimized for long-context applications. Our vertically integrated pipeline unites domain-balanced, frequency-aware data curation; multi-stage pre-training with quality annealing and length extension to 128k tokens; and targeted post-training via supervised fine-tuning, preference learning, and online reinforcement learning. xGen-small delivers strong performance across various tasks, especially in math and coding domains, while excelling at long context benchmarks.

The rapid development of Internet of Things (IoT) technology has transformed people's way of life and has a profound impact on both production and daily activities. However, with the rapid advancement of IoT technology, the security of IoT devices has become an unavoidable issue in both research and applications. Although some efforts have been made to detect or mitigate IoT security vulnerabilities, they often struggle to adapt to the complexity of IoT environments, especially when dealing with dynamic security scenarios. How to automatically, efficiently, and accurately understand these vulnerabilities remains a challenge. To address this, we propose an IoT security assistant driven by Large Language Model (LLM), which enhances the LLM's understanding of IoT security vulnerabilities and related threats. The aim of the ICoT method we propose is to enable the LLM to understand security issues by breaking down the various dimensions of security vulnerabilities and generating responses tailored to the user's specific needs and expertise level. By incorporating ICoT, LLM can gradually analyze and reason through complex security scenarios, resulting in more accurate, in-depth, and personalized security recommendations and solutions. Experimental results show that, compared to methods relying solely on LLM, our proposed LLM-driven IoT security assistant significantly improves the understanding of IoT security issues through the ICoT approach and provides personalized solutions based on the user's identity, demonstrating higher accuracy and reliability.