--Retrieval-Augmented Generation (RAG) has become a standard approach for improving the reliability of large language models (LLMs). Prior work demonstrates the vulnerability of RAG systems by misleading them into generating attacker-chosen outputs through poisoning the knowledge base. However, this paper uncovers that such attacks could be mitigated by the strong self-correction ability (SCA) of modern LLMs, which can reject false context once properly configured. This SCA poses a significant challenge for attackers aiming to manipulate RAG systems. RAG, a new poisoning paradigm that compromises the retriever itself to suppress the SCA and enforce attacker-chosen outputs. This compromisation enables the attacker to straightforwardly embed anti-SCA instructions into the context provided to the generator, thereby bypassing the SCA. T o this end, we present a contrastive-learning-based model editing technique that performs localized and stealthy edits, ensuring the retriever returns a malicious instruction only for specific victim queries while preserving benign retrieval behavior . T o further strengthen the attack, we design an iterative co-optimization framework that automatically discovers robust instructions capable of bypassing prompt-based defenses. We extensively evaluate DisarmRAG across six LLMs and three QA benchmarks. Our results show near-perfect retrieval of malicious instructions, which successfully suppress SCA and achieve attack success rates exceeding 90% under diverse defensive prompts. Also, the edited retriever remains stealthy under several detection methods, highlighting the urgent need for retriever-centric defenses. Modern large language models (LLMs) achieve remarkable performance across a wide range of tasks [32], [26], [38]. Despite their success, LLMs are also well known for their hallucination behaviors [25], which generate fabricated content. Such unreliability limits their deployment in critical domains, including healthcare [69] and law [10]. Retrieval-augmented generation (RAG) [37], [29] has emerged as a promising paradigm to mitigate these limitations. By integrating external knowledge, RAG enables LLMs to generate more reliable responses. A key component of RAG is the retriever [27], which encodes both user queries and documents from an external knowledge base [72], [11]. The retriever identifies documents that are most relevant to the input query. These retrieved documents are then combined with the query to guide the LLM in producing grounded responses. Although RAG systems enhance LLMs with external knowledge, their deployment introduces new attack surfaces. Prior work [84], [81], [41], [6] demonstrates the effectiveness of misleading the system to give attack-chosen outputs through injecting malicious content into the knowledge base.

For human cognitive process, spatial reasoning and perception are closely entangled, yet the nature of this interplay remains underexplored in the evaluation of multimodal large language models (MLLMs). While recent MLLM advancements show impressive performance on reasoning, their capacity for human-like spatial cognition remains an open question. In this work, we introduce a systematic evaluation framework to assess the spatial reasoning abilities of state-of-the-art MLLMs relative to human performance. Central to our work is 11Plus-Bench, a high-quality benchmark derived from realistic standardized spatial aptitude tests. 11Plus-Bench also features fine-grained expert annotations of both perceptual complexity and reasoning process, enabling detailed instance-level analysis of model behavior. Through extensive experiments across 14 MLLMs and human evaluation, we find that current MLLMs exhibit early signs of spatial cognition. Despite a large performance gap compared to humans, MLLMs' cognitive profiles resemble those of humans in that cognitive effort correlates strongly with reasoning-related complexity. However, instance-level performance in MLLMs remains largely random, whereas human correctness is highly predictable and shaped by abstract pattern complexity. These findings highlight both emerging capabilities and limitations in current MLLMs' spatial reasoning capabilities and provide actionable insights for advancing model design.

Age-related Macular Degeneration (AMD) is a prevalent eye condition affecting visual acuity. Anti-vascular endothelial growth factor (anti-VEGF) treatments have been effective in slowing the progression of neovascular AMD, with better outcomes achieved through timely diagnosis and consistent monitoring. Tracking the progression of neovascular activity in OCT scans of patients with exudative AMD allows for the development of more personalized and effective treatment plans. This was the focus of the Monitoring Age-related Macular Degeneration Progression in Optical Coherence Tomography (MARIO) challenge, in which we participated. In Task 1, which involved classifying the evolution between two pairs of 2D slices from consecutive OCT acquisitions, we employed a fusion CNN network with model ensembling to further enhance the model's performance. For Task 2, which focused on predicting progression over the next three months based on current exam data, we proposed the Patch Progression Masked Autoencoder that generates an OCT for the next exam and then classifies the evolution between the current OCT and the one generated using our solution from Task 1. The results we achieved allowed us to place in the Top 10 for both tasks. Some team members are part of the same organization as the challenge organizers; therefore, we are not eligible to compete for the prize.

The growing adoption of foundation models calls for a paradigm shift from Data Science to Model Science. Unlike data-centric approaches, Model Science places the trained model at the core of analysis, aiming to interact, verify, explain, and control its behavior across diverse operational contexts. This paper introduces a conceptual framework for a new discipline called Model Science, along with the proposal for its four key pillars: V erification, which requires strict, context-aware evaluation protocols; Explanation, which is understood as various approaches to explore of internal model operations; Control, which integrates alignment techniques to steer model behavior; and Interface, which develops interactive and visual explanation tools to improve human calibration and decision-making. The proposed framework aims to guide the development of credible, safe, and human-aligned AI systems.

The paper presents a visio-verbal teleimpedance interface for commanding 3D stiffness ellipsoids to the remote robot with a combination of the operator's gaze and verbal interaction. The gaze is detected by an eye-tracker, allowing the system to understand the context in terms of what the operator is currently looking at in the scene. Along with verbal interaction, a Visual Language Model (VLM) processes this information, enabling the operator to communicate their intended action or provide corrections. Based on these inputs, the interface can then generate appropriate stiffness matrices for different physical interaction actions. To validate the proposed visio-verbal teleimpedance interface, we conducted a series of experiments on a setup including a Force Dimension Sigma.7 haptic device to control the motion of the remote Kuka LBR iiwa robotic arm. The human operator's gaze is tracked by Tobii Pro Glasses 2, while human verbal commands are processed by a VLM using GPT-4o. The first experiment explored the optimal prompt configuration for the interface. The second and third experiments demonstrated different functionalities of the interface on a slide-in-the-groove task.

The ability to research and synthesize knowledge is central to human expertise and progress. An emerging class of systems promises these exciting capabilities through generative research synthesis, performing retrieval over the live web and synthesizing discovered sources into long-form, cited summaries. However, evaluating such systems remains an open challenge: existing question-answering benchmarks focus on short-form factual responses, while expert-curated datasets risk staleness and data contamination. Both fail to capture the complexity and evolving nature of real research synthesis tasks. In this work, we introduce DeepScholar-bench, a live benchmark and holistic, automated evaluation framework designed to evaluate generative research synthesis. DeepScholar-bench draws queries from recent, high-quality ArXiv papers and focuses on a real research synthesis task: generating the related work sections of a paper by retrieving, synthesizing, and citing prior research. Our evaluation framework holistically assesses performance across three key dimensions, knowledge synthesis, retrieval quality, and verifiability. We also develop DeepScholar-base, a reference pipeline implemented efficiently using the LOTUS API. Using the DeepScholar-bench framework, we perform a systematic evaluation of prior open-source systems, search AI's, OpenAI's DeepResearch, and DeepScholar-base. We find that DeepScholar-base establishes a strong baseline, attaining competitive or higher performance than each other method. We also find that DeepScholar-bench remains far from saturated, with no system exceeding a score of $19\%$ across all metrics. These results underscore the difficulty of DeepScholar-bench, as well as its importance for progress towards AI systems capable of generative research synthesis. We make our code available at https://github.com/guestrin-lab/deepscholar-bench.

Backdoor attacks are a significant threat to the performance and integrity of pre-trained language models. Although such models are routinely fine-tuned for downstream NLP tasks, recent work shows they remain vulnerable to backdoor attacks that survive vanilla fine-tuning. These attacks are difficult to defend because end users typically lack knowledge of the attack triggers. Such attacks consist of stealthy malicious triggers introduced through subtle syntactic or stylistic manipulations, which can bypass traditional detection and remain in the model, making post-hoc purification essential. In this study, we explore whether attention-head pruning can mitigate these threats without any knowledge of the trigger or access to a clean reference model. To this end, we design and implement six pruning-based strategies: (i) gradient-based pruning, (ii) layer-wise variance pruning, (iii) gradient-based pruning with structured L1/L2 sparsification, (iv) randomized ensemble pruning, (v) reinforcement-learning-guided pruning, and (vi) Bayesian uncertainty pruning. Each method iteratively removes the least informative heads while monitoring validation accuracy to avoid over-pruning. Experimental evaluation shows that gradient-based pruning performs best while defending the syntactic triggers, whereas reinforcement learning and Bayesian pruning better withstand stylistic attacks.

With the growing complexity of modern integrated circuits, hardware engineers are required to devote more effort to the full design-to-manufacturing workflow. This workflow involves numerous iterations, making it both labor-intensive and error-prone. Therefore, there is an urgent demand for more efficient Electronic Design Automation (EDA) solutions to accelerate hardware development. Recently, large language models (LLMs) have shown remarkable advancements in contextual comprehension, logical reasoning, and generative capabilities. Since hardware designs and intermediate scripts can be represented as text, integrating LLM for EDA offers a promising opportunity to simplify and even automate the entire workflow. Accordingly, this paper provides a comprehensive overview of incorporating LLMs into EDA, with emphasis on their capabilities, limitations, and future opportunities. Three case studies, along with their outlook, are introduced to demonstrate the capabilities of LLMs in hardware design, testing, and optimization. Finally, future directions and challenges are highlighted to further explore the potential of LLMs in shaping the next-generation EDA, providing valuable insights for researchers interested in leveraging advanced AI technologies for EDA.

Cataract surgery remains one of the most widely performed and effective procedures for vision restoration. Effective surgical planning requires integrating diverse clinical examinations for patient assessment, intraocular lens (IOL) selection, and risk evaluation. Large language models (LLMs) have shown promise in supporting clinical decision-making. However, existing LLMs often lack the domain-specific expertise to interpret heterogeneous ophthalmic data and provide actionable surgical plans. To enhance the model's ability to interpret heterogeneous ophthalmic reports, we propose a knowledge-driven Multi-Agent System (MAS), where each agent simulates the reasoning process of specialist ophthalmologists, converting raw clinical inputs into structured, actionable summaries in both training and deployment stages. Building on MAS, we introduce CataractSurg-80K, the first large-scale benchmark for cataract surgery planning that incorporates structured clinical reasoning. Each case is annotated with diagnostic questions, expert reasoning chains, and structured surgical recommendations. We further introduce Qwen-CSP, a domain-specialized model built on Qwen-4B, fine-tuned through a multi-stage process tailored for surgical planning. Comprehensive experiments show that Qwen-CSP outperforms strong general-purpose LLMs across multiple metrics. Our work delivers a high-quality dataset, a rigorous benchmark, and a domain-adapted LLM to facilitate future research in medical AI reasoning and decision support.

Fine-tuning multi-turn dialogue systems requires high-quality supervision but often suffers from degraded performance when exposed to low-quality data. Supervision errors in early turns can propagate across subsequent turns, undermining coherence and response quality. Existing methods typically address data quality via static prefiltering, which decouples quality control from training and fails to mitigate turn-level error propagation. In this context, we propose ReSURE (Regularizing Supervision UnREliability), an adaptive learning method that dynamically down-weights unreliable supervision without explicit filtering. ReSURE estimates per-turn loss distributions using Welford's online statistics and reweights sample losses on the fly accordingly. Experiments on both single-source and mixed-quality datasets show improved stability and response quality. Notably, ReSURE enjoys positive Spearman correlations (0.21 ~ 1.0 across multiple benchmarks) between response scores and number of samples regardless of data quality, which potentially paves the way for utilizing large-scale data effectively. Code is publicly available at https://github.com/Elvin-Yiming-Du/ReSURE_Multi_Turn_Training.