Archives AI News

arXiv:2508.19851v1 Announce Type: new Abstract: Large Language Models (LLMs) exhibit emergent capabilities in structured domains, suggesting they may implicitly internalize high-fidelity representations of world models. While probing techniques have shown promising signs of this in scientific and game-based settings, they rely on model-specific internal activations, which limit interpretability and generalizability. In this work, we propose a model-agnostic, state-based evaluation framework using chess as a benchmark to assess whether LLMs preserve the semantics of structured environments. Our method analyzes the downstream legal move distributions (state affordances) to estimate semantic fidelity between predicted and actual game states. This approach offers a more meaningful evaluation than conventional string-based metrics by aligning more closely with the strategic and rule-governed nature of chess. Experimental results demonstrate that our metrics capture deficiencies in state-tracking, highlighting limitations of LLMs in maintaining coherent internal models over long sequences. Our framework provides a robust tool for evaluating structured reasoning in LLMs without requiring internal model access, and generalizes to a wide class of symbolic environments.

arXiv:2505.11717v2 Announce Type: replace-cross Abstract: Multi-modal large language model (MLLM)-based web agents interact with webpage environments by generating actions based on screenshots of the webpages. Environmental prompt injection attacks manipulate the environment to induce the web agent to perform a specific, attacker-chosen action--denoted as the target action. However, existing attacks suffer from limited effectiveness or stealthiness, or are impractical in real-world settings. In this work, we propose EnvInjection, a new attack that addresses these limitations. Our attack adds a perturbation to the raw pixel values of the rendered webpage. After these perturbed pixels are mapped into a screenshot, the perturbation induces the web agent to perform the target action. We formulate the task of finding the perturbation as an optimization problem. A key challenge in solving this problem is that the mapping between raw pixel values and screenshot is non-differentiable, making it difficult to backpropagate gradients to the perturbation. To overcome this, we train a neural network to approximate the mapping and apply projected gradient descent to solve the reformulated optimization problem. Extensive evaluation on multiple webpage datasets shows that EnvInjection is highly effective and significantly outperforms existing baselines.

arXiv:2508.19932v1 Announce Type: new Abstract: The proliferation of digital payment platforms has transformed commerce, offering unmatched convenience and accessibility globally. However, this growth has also attracted malicious actors, leading to a corresponding increase in sophisticated social engineering scams. These scams are often initiated and orchestrated on multiple surfaces outside the payment platform, making user and transaction-based signals insufficient for a complete understanding of the scam's methodology and underlying patterns, without which it is very difficult to prevent it in a timely manner. This paper presents CASE (Conversational Agent for Scam Elucidation), a novel Agentic AI framework that addresses this problem by collecting and managing user scam feedback in a safe and scalable manner. A conversational agent is uniquely designed to proactively interview potential victims to elicit intelligence in the form of a detailed conversation. The conversation transcripts are then consumed by another AI system that extracts information and converts it into structured data for downstream usage in automated and manual enforcement mechanisms. Using Google's Gemini family of LLMs, we implemented this framework on Google Pay (GPay) India. By augmenting our existing features with this new intelligence, we have observed a 21% uplift in the volume of scam enforcements. The architecture and its robust evaluation framework are highly generalizable, offering a blueprint for building similar AI-driven systems to collect and manage scam intelligence in other sensitive domains.

arXiv:2508.19963v1 Announce Type: new Abstract: Optimizing modern production plants using the job-shop principle is a known hard problem. For very large plants, like semiconductor fabs, the problem becomes unsolvable on a plant-wide scale in a reasonable amount of time using classical linear optimization. An alternative approach is the use of swarm intelligence algorithms. These have been applied to the job-shop problem before, but often in a centrally calculated way where they are applied to the solution space, but they can be implemented in a bottom-up fashion to avoid global result computation as well. One of the problems in semiconductor production is that the production process requires a lot of switching between machines that process lots one after the other and machines that process batches of lots at once, often with long processing times. In this paper, we address this switching problem with the ``boids'' flocking algorithm that was originally used in robotics and movie industry. The flocking behavior is a bio-inspired algorithm that uses only local information and interaction based on simple heuristics. We show that this algorithm addresses these valid considerations in production plant optimization, as it reacts to the switching of machine kinds similar to how a swarm of flocking animals would react to obstacles in its course.

arXiv:2508.16439v3 Announce Type: replace-cross Abstract: Large language models (LLMs) and vision-augmented LLMs (VLMs) have significantly advanced medical informatics, diagnostics, and decision support. However, these models exhibit systematic biases, particularly age bias, compromising their reliability and equity. This is evident in their poorer performance on pediatric-focused text and visual question-answering tasks. This bias reflects a broader imbalance in medical research, where pediatric studies receive less funding and representation despite the significant disease burden in children. To address these issues, a new comprehensive multi-modal pediatric question-answering benchmark, PediatricsMQA, has been introduced. It consists of 3,417 text-based multiple-choice questions (MCQs) covering 131 pediatric topics across seven developmental stages (prenatal to adolescent) and 2,067 vision-based MCQs using 634 pediatric images from 67 imaging modalities and 256 anatomical regions. The dataset was developed using a hybrid manual-automatic pipeline, incorporating peer-reviewed pediatric literature, validated question banks, existing benchmarks, and existing QA resources. Evaluating state-of-the-art open models, we find dramatic performance drops in younger cohorts, highlighting the need for age-aware methods to ensure equitable AI support in pediatric care.

arXiv:2508.20018v1 Announce Type: new Abstract: The rapid advancement of large vision language models (LVLMs) and agent systems has heightened interest in mobile GUI agents that can reliably translate natural language into interface operations. Existing single-agent approaches, however, remain limited by structural constraints. Although multi-agent systems naturally decouple different competencies, recent progress in multi-agent reinforcement learning (MARL) has often been hindered by inefficiency and remains incompatible with current LVLM architectures. To address these challenges, we introduce SWIRL, a staged workflow for interleaved reinforcement learning designed for multi-agent systems. SWIRL reformulates MARL into a sequence of single-agent reinforcement learning tasks, updating one agent at a time while keeping the others fixed. This formulation enables stable training and promotes efficient coordination across agents. Theoretically, we provide a stepwise safety bound, a cross-round monotonic improvement theorem, and convergence guarantees on return, ensuring robust and principled optimization. In application to mobile GUI control, SWIRL instantiates a Navigator that converts language and screen context into structured plans, and an Interactor that grounds these plans into executable atomic actions. Extensive experiments demonstrate superior performance on both high-level and low-level GUI benchmarks. Beyond GUI tasks, SWIRL also demonstrates strong capability in multi-agent mathematical reasoning, underscoring its potential as a general framework for developing efficient and robust multi-agent systems.

arXiv:2508.19697v1 Announce Type: cross Abstract: Current safety alignment for large language models(LLMs) continues to present vulnerabilities, given that adversarial prompting can effectively bypass their safety measures.Our investigation shows that these safety mechanisms predominantly depend on a limited subset of attention heads: removing or ablating these heads can severely compromise model safety. To identify and evaluate these safety-critical components, we introduce RDSHA, a targeted ablation method that leverages the model's refusal direction to pinpoint attention heads mostly responsible for safety behaviors. Further analysis shows that existing jailbreak attacks exploit this concentration by selectively bypassing or manipulating these critical attention heads. To address this issue, we propose AHD, a novel training strategy designed to promote the distributed encoding of safety-related behaviors across numerous attention heads. Experimental results demonstrate that AHD successfully distributes safety-related capabilities across more attention heads. Moreover, evaluations under several mainstream jailbreak attacks show that models trained with AHD exhibit considerably stronger safety robustness, while maintaining overall functional utility.

arXiv:2508.20040v1 Announce Type: new Abstract: 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: Verification, 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.

arXiv:2508.19839v1 Announce Type: cross Abstract: Model merging has emerged as an efficient strategy for constructing multitask models by integrating the strengths of multiple available expert models, thereby reducing the need to fine-tune a pre-trained model for all the tasks from scratch. Existing data-independent methods struggle with performance limitations due to the lack of data-driven guidance. Data-driven approaches also face key challenges: gradient-based methods are computationally expensive, limiting their practicality for merging large expert models, whereas existing gradient-free methods often fail to achieve satisfactory results within a limited number of optimization steps. To address these limitations, this paper introduces PSO-Merging, a novel data-driven merging method based on the Particle Swarm Optimization (PSO). In this approach, we initialize the particle swarm with a pre-trained model, expert models, and sparsified expert models. We then perform multiple iterations, with the final global best particle serving as the merged model. Experimental results on different language models show that PSO-Merging generally outperforms baseline merging methods, offering a more efficient and scalable solution for model merging.

arXiv:2508.19026v1 Announce Type: cross Abstract: This paper introduces MovieCORE, a novel video question answering (VQA) dataset designed to probe deeper cognitive understanding of movie content. Unlike existing datasets that focus on surface-level comprehension, MovieCORE emphasizes questions that engage System-2 thinking while remaining specific to the video material. We present an innovative agentic brainstorming approach, utilizing multiple large language models (LLMs) as thought agents to generate and refine high-quality question-answer pairs. To evaluate dataset quality, we develop a set of cognitive tests assessing depth, thought-provocation potential, and syntactic complexity. We also propose a comprehensive evaluation scheme for assessing VQA model performance on deeper cognitive tasks. To address the limitations of existing video-language models (VLMs), we introduce an agentic enhancement module, Agentic Choice Enhancement (ACE), which improves model reasoning capabilities post-training by up to 25%. Our work contributes to advancing movie understanding in AI systems and provides valuable insights into the capabilities and limitations of current VQA models when faced with more challenging, nuanced questions about cinematic content. Our project page, dataset and code can be found at https://joslefaure.github.io/assets/html/moviecore.html.