Recent advances in AI-assisted education have encouraged the integration of vision-language models (VLMs) into academic assessment, particularly for tasks that require both quantitative and qualitative evaluation. However, existing VLM based approaches struggle with complex educational artifacts, such as programming tasks with executable components and measurable outputs, that require structured reasoning and alignment with clearly defined evaluation criteria. We introduce AGACCI, a multi-agent system that distributes specialized evaluation roles across collaborative agents to improve accuracy, interpretability, and consistency in code-oriented assessment. To evaluate the framework, we collected 360 graduate-level code-based assignments from 60 participants, each annotated by domain experts with binary rubric scores and qualitative feedback. Experimental results demonstrate that AGACCI outperforms a single GPT-based baseline in terms of rubric and feedback accuracy, relevance, consistency, and coherence, while preserving the instructional intent and evaluative depth of expert assessments. Although performance varies across task types, AGACCI highlights the potential of multi-agent systems for scalable and context-aware educational evaluation.

The rapid advancements of AI agents have ignited the long-held ambition of leveraging them to accelerate scientific discovery. Achieving this goal requires a deep understanding of the frontiers of human knowledge. As such, Humanity's Last Exam (HLE) provides an exceptionally challenging touchstone for evaluating scientific AI agents. In this work, we aim to construct the foundational architecture for general-purpose agents and validate the capabilities through leading performance on HLE. To achieve this, we introduce X-Master, a tool-augmented reasoning agent designed to emulate human researchers by interacting flexibly with external tools during its reasoning process. This agent, guided by the conceptualization of code as an interaction language, can flexibly leverage built-in Python libraries and our customized tools to augment the reasoning. We further scale its capabilities through X-Masters, a scattered-and-stacked agentic workflow that systematically enhances breadth and depth of reasoning. Our open-source solution, X-Masters, sets a new state-of-the-art record on HLE with a score of 32.1%, surpassing OpenAI's and Google's Deep Research (26.6% and 26.9%) and becoming the first to exceed the 30% threshold. This work allows us to gain a deeper understanding of complex task-solving and accumulates valuable experience that can inform future advancements, guiding subsequent model training.

High-Frequency Trading (HFT) is pivotal in cryptocurrency markets, demanding rapid decision-making. Social media platforms like Reddit offer valuable, yet underexplored, information for such high-frequency, short-term trading. This paper introduces \textbf{PulseReddit}, a novel dataset that is the first to align large-scale Reddit discussion data with high-frequency cryptocurrency market statistics for short-term trading analysis. We conduct an extensive empirical study using Large Language Model (LLM)-based Multi-Agent Systems (MAS) to investigate the impact of social sentiment from PulseReddit on trading performance. Our experiments conclude that MAS augmented with PulseReddit data achieve superior trading outcomes compared to traditional baselines, particularly in bull markets, and demonstrate robust adaptability across different market regimes. Furthermore, our research provides conclusive insights into the performance-efficiency trade-offs of different LLMs, detailing significant considerations for practical model selection in HFT applications. PulseReddit and our findings establish a foundation for advanced MAS research in HFT, demonstrating the tangible benefits of integrating social media.

The increasing deployment of Artificial Intelligence (AI) and other autonomous algorithmic systems presents the world with new systemic risks. While focus often lies on the function of individual algorithms, a critical and underestimated danger arises from their interactions, particularly when algorithmic systems operate without awareness of each other, or when those deploying them are unaware of the full algorithmic ecosystem deployment is occurring in. These interactions can lead to unforeseen, rapidly escalating negative outcomes - from market crashes and energy supply disruptions to potential physical accidents and erosion of public trust - often exceeding the human capacity for effective monitoring and the legal capacities for proper intervention. Current governance frameworks are inadequate as they lack visibility into this complex ecosystem of interactions. This paper outlines the nature of this challenge and proposes some initial policy suggestions centered on increasing transparency and accountability through phased system registration, a licensing framework for deployment, and enhanced monitoring capabilities.

--Traffic congestion remains a major challenge for modern urban transportation, diminishing both efficiency and quality of life. While autonomous driving technologies and reinforcement learning (RL) have shown promise for improving traffic control, most prior work has focused on small-scale networks or isolated intersections. Large-scale mixed traffic control, involving both human-driven and robotic vehicles, remains underexplored. In this study, we propose a decentralized multi-agent reinforcement learning framework for managing large-scale mixed traffic networks, where intersections are controlled either by traditional traffic signals or by robotic vehicles. We evaluate our approach on a real-world network of 14 intersections in Colorado Springs, Colorado, USA, using average vehicle waiting time as the primary measure of traffic efficiency. We are exploring a problem that has not been sufficiently addressed: Is large-scale Multi-Agent Traffic Control (MTC) still feasible when facing time-varying Origin-Destination (OD) patterns?

Cyber-physical systems (CPS) designed in simulators, often consisting of multiple interacting agents (e.g. in multi-agent formations), behave differently in the real-world. We want to verify these systems during runtime when they are deployed. We thus propose robust predictive runtime verification (RPRV) algorithms for: (1) general stochastic CPS under signal temporal logic (STL) tasks, and (2) stochastic multi-agent systems (MAS) under spatio-temporal logic tasks. The RPRV problem presents the following challenges: (1) there may not be sufficient data on the behavior of the deployed CPS, (2) predictive models based on design phase system trajectories may encounter distribution shift during real-world deployment, and (3) the algorithms need to scale to the complexity of MAS and be applicable to spatio-temporal logic tasks. To address the challenges, we assume knowledge of an upper bound on the statistical distance between the trajectory distributions of the system at deployment and design time. We are motivated by our prior work [1, 2] where we proposed an accurate and an interpretable RPRV algorithm for general CPS, which we here extend to the MAS setting and spatio-temporal logic tasks. Specifically, we use a learned predictive model to estimate the system behavior at runtime and robust conformal prediction to obtain probabilistic guarantees by accounting for distribution shifts. Building on [1], we perform robust conformal prediction over the robust semantics of spatio-temporal reach and escape logic (STREL) to obtain centralized RPRV algorithms for MAS. We empirically validate our results in a drone swarm simulator, where we show the scalability of our RPRV algorithms to MAS and analyze the impact of different trajectory predictors on the verification result. To the best of our knowledge, these are the first statistically valid algorithms for MAS under distribution shift.

Scoring rules elicit probabilistic predictions from a strategic agent by scoring the prediction against a ground truth state. A scoring rule is proper if, from the agent's perspective, reporting the true belief maximizes the expected score. With the development of language models, Wu and Hartline (2024) proposes a reduction from textual information elicitation to the numerical (i.e. probabilistic) information elicitation problem, which achieves provable properness for textual elicitation. However, not all proper scoring rules are well aligned with human preference over text. Our paper designs the Aligned Scoring rule (ASR) for text by optimizing and minimizing the mean squared error between a proper scoring rule and a reference score (e.g. human score). Our experiments show that our ASR outperforms previous methods in aligning with human preference while maintaining properness.

Recent advances in AI agents capable of solving complex, everyday tasks, from scheduling to customer service, have enabled deployment in real-world settings, but their possibilities for unsafe behavior demands rigorous evaluation. While prior benchmarks have attempted to assess agent safety, most fall short by relying on simulated environments, narrow task domains, or unrealistic tool abstractions. We introduce OpenAgentSafety, a comprehensive and modular framework for evaluating agent behavior across eight critical risk categories. Unlike prior work, our framework evaluates agents that interact with real tools, including web browsers, code execution environments, file systems, bash shells, and messaging platforms; and supports over 350 multi-turn, multi-user tasks spanning both benign and adversarial user intents. OpenAgentSafety is designed for extensibility, allowing researchers to add tools, tasks, websites, and adversarial strategies with minimal effort. It combines rule-based analysis with LLM-as-judge assessments to detect both overt and subtle unsafe behaviors. Empirical analysis of five prominent LLMs in agentic scenarios reveals unsafe behavior in 51.2% of safety-vulnerable tasks with Claude-Sonnet-3.7, to 72.7% with o3-mini, highlighting critical safety vulnerabilities and the need for stronger safeguards before real-world deployment.

-- In disaster response or surveillance operations, quickly identifying areas needing urgent attention is critical, but deploying response teams to every location is inefficient or often impossible. Effective performance in this domain requires coordinating a multi-robot inspection team to prioritize inspecting locations more likely to need immediate response, while also minimizing travel time. This is particularly challenging because robots must directly observe the locations to determine which ones require additional attention. This work introduces a multi-robot planning framework for coordinated time-critical multi-robot search under uncertainty. Our approach uses a graph neural network to estimate the likelihood of PoIs needing attention from noisy sensor data and then uses those predictions to guide a multi-robot model-based planner to determine the cost-effective plan. Simulated experiments demonstrate that our planner improves performance at least by 16.3%, 26.7%, and 26.2% for 1, 3, and 5 robots, respectively, compared to non-learned and learned baselines. In scenarios like disaster aftermath inspection or critical surveillance operations, quickly traveling to and inspecting affected areas is crucial for an efficient response.

Continual Multi-Agent Reinforcement Learning (Co-MARL) requires agents to address catastrophic forgetting issues while learning new coordination policies with the dynamics team. In this paper, we delve into the core of Co-MARL, namely Relation Patterns, which refer to agents' general understanding of interactions. In addition to generality, relation patterns exhibit task-specificity when mapped to different action spaces. To this end, we propose a novel method called General Relation Patterns-Guided Task-Specific Decision-Maker (RPG). In RPG, agents extract relation patterns from dynamic observation spaces using a relation capturer. These task-agnostic relation patterns are then mapped to different action spaces via a task-specific decision-maker generated by a conditional hypernetwork. To combat forgetting, we further introduce regularization items on both the relation capturer and the conditional hypernetwork. Results on SMAC and LBF demonstrate that RPG effectively prevents catastrophic forgetting when learning new tasks and achieves zero-shot generalization to unseen tasks.