CATArena Advances AI Agent Testing While Denario Simplifies Financial Research

Key Takeaways

• Certified runtime alarms for computer-use agents can improve task accuracy and reduce failures.
• Retrieving relations and detecting fallacies in political debates can improve the accuracy of fallacy detection and classification.
• Training communication-efficient mixture-of-experts language models can improve the performance of language models and reduce computational costs.
• Object-centric predictive monitoring of collaborative processes can improve the accuracy of predictive monitoring and reduce the risk of errors.
• Ultra-concise bullet points highlighting absolute critical findings can improve the clarity and effectiveness of AI research.
• The importance of certified runtime alarms and more accurate and reliable language models has been emphasized.
• The potential for AI to improve human-centered explainable AI has been highlighted.
• Training GUI agents to reason and leverage world models with reinforcement learning can improve the performance of GUI agents.
• The need for more accurate and reliable language models has been emphasized.
• The importance of certified runtime alarms and more accurate and reliable language models has been highlighted.

Sources

• CURA: Certified Runtime Alarms for Computer-Use Agents
• Retrieving Relations, Detecting Fallacies: A RAG Approach to Political Debate Analysis
• Not All Explanations Are Sought: Information-Seeking Psychology for Human-Centered XAI
• Rating the Raters: Rasch Measurement Theory for LLM Evaluation
• Agents for Everyone: A Workshop Framework for Building Agentic AI Capabilities in a Distributed Curation Community
• A Framework for Object-Centric Predictive Monitoring of Collaborative Processes
• If Agents Were Angels, No Governance Would Be Necessary: Out-of-Band Policy Enforcement at a Trusted Tool Boundary
• ReToolSQL: Agentic Reinforcement Learning for Robust Text-to-SQL
• Why Didn't It Check? Unsupported Final Claims and Their Repair in Two Tool-Equipped Language Models
• PhenoIntel: A Lifecycle-Aligned Multi-Agent Web Application for Verified, Accessible Plant Phenotype Analysis
• Prove2Me: An Open Collaborative Platform for Scaling Math Formalization
• Program Learning with Verifiable Rewards: Symbolic Backpropagation for Post-Training LLMs
• VERA-8B: Evidence-Grounded Audit Risk Reasoning from SEC Filings
• RetailAgent: Structured Adverse Timing in Self-Conditioned Multimodal LLM Trading Agents
• Logos: An Agent Harness on a Cross-Process Bus
• InstructMesh: Selective Refinement of Generative 3D Models for Fabrication
• When Robots Mishear Us: Mapping the Safety Risks of Voice-Controlled Embodied AI
• Training Communication-Efficient Mixture-of-Experts Language Models with Layer Re-Configuration
• AcrossVAM1.0: Particle World Modeling for Text-Assisted Robot Video Prediction
• COVER: Identifiable Evaluation of Coalition Routing
• SABER: Stability-Aware Early Exit for LLM Reasoning via Adversarial Branch Probing
• CASTANET: Causality-Aware Spatio-Temporal Adversarial Network Using Traffic Incident Effects
• Generative AI Expands the Intellectual Reach of Course Based Undergraduate Research Experiences (CUREs)
• CareGraph: An Auditable Hybrid AI Framework for Evidence-Grounded Personalized Longitudinal Health Intelligence
• Effectiveness of IoT and Deep Learning for Detection and Severity Assessment of Postelectrotermes militaris in Tea Plantations
• Benchmarking General Mobile Assistants in Challenging Real-World Scenarios
• Class-Based Heuristic Selection for Solving the Flying Block Puzzle
• Hypothesize, Evaluate, Refine: A Scientific Agent for PDE Discovery with Unknown Spatial Coefficient Fields
• SETU: An Agentic Ecosystem for Multilingual, Persona-Aware Communication Coaching
• WM-R1: Training GUI Agents to Reason and leverage World Models with Reinforcement Learning
• Thinking Costs Tokens: When More Structure is Worth the Price
• LongGuard: Mechanistic Analysis and Training-Free Mitigation of Long-Context Failure in Safety Guardrails
• A Deep Learning-Based Stacking Ensemble Framework for Turbofan Engine Remaining Useful Life Prediction
• From Documents to Reasoning: A Validated Synthetic Data Pipeline and Semantic-Aware Fine-Tuning for Financial Numerical Reasoning
• AI Alignment through a Game-theoretic Lens: A Survey
• Rubric-to-Code Credit Assignment for Reinforcement Learning
• When Teacher Guidance Misleads: Reward-Aligned On-Policy Distillation
• Cross-Session Decomposition Attacks: Scaling Risk and Intent-Aligned Retrieval Defense
• Stay Within Your Bounds: Distance-Guided Decoding for Guaranteed Context-Free Grammar Compliance
• Generative AI Alignment with Hinduism's Theological Plurality and Sacred Representation
• Expert Knowledge & Machine Understanding: Bridging Reactome's Ontology with LLM Semantic Embeddings
• CrabOS: An Operating System for Human-AI Co-inhabitation
• Physics-Guided Flow Matching for CT Image Reconstruction
• Beyond Task-Only Matching: Personalized Skill Routing with Counterfactual Evaluation
• The Illusion of $\textit{What If}$: Evaluating the Breakdown of Counterfactual Reasoning in LLMs
• CoRe-MoE: Compact Reusable MoE for Continual Multimodal Instruction Tuning
• From Uncertainty to Clinical Risk: Severity-Aware Conformal Planning for Interactive Medical Diagnosis
• An Empirical Evaluation of Cross-City POI Recommendation on a Large-Scale Benchmark
• Evidential-Based Higher-Order Set Argumentation Framework
• AcCoRD: Evaluating User-Agent Collaboration Under Realistic User Preference Dynamics
• CEDAR: Automata as Verifiable Interfaces for Language-Guided Embodied Action
• Nemotron 3.5 Content Safety Moderator: A Compact Multimodal, Multilingual, and Reasoning Enabled Content Safety Moderator
• Context Localization for Generalized Level-Based Evaluation in Knowledge-Based Systems
• Under-Mattress Temporal Sensing for Next-Day Agitation Risk Scoring in Dementia Wards
• Speculative Probing: LLM Monitoring at Speculative-Decoding Cost
• Learning to Allocate Incentives for Incentivized Advertising via Offline Model-Based Reinforcement Learning
• Timing-Aware Repurchase Prediction for Web-Scale E-Commerce: Survival Models for Multi-Surface Grocery Recommendation
• LLM-Augmented Causal Discovery: Probabilistic Fusion of Edge Existence and Orientation
• Time Capsule of Testable Human Knowledge: 41 Years of Jeopardy! in a Single Free Local Model
• KLOD: Locality-Preserving Knowledge Editing via Non-Target Distribution Preservation
• RealSWE: A Compositional Evaluation of Coding Agents under Realistic User Requests
• String: An Agentic OS Where Every App Is a Markdown File
• Coverage, Not Credit: Failure-Credit Routing of Zeroth-Order Perturbation Budgets Does Not Improve On-Pool Sample Efficiency for LLM Agents
• SpikeOPD: Stable On-Policy Distillation for Autoregressive Spiking Language Models
• WeAgent-MMSearch: Native Text-Vision Interaction for Multimodal Search Agents
• See, Hypothesize, Validate: Multimodal Agentic Framework for Discovering Governing PDEs
• Resource Constraints and Performance in Agentic AI Systems
• Should I Use This Synthetic Dataset for Training? How to Test with Minimal Real Data
• When Evidence Shapes Collaboration: Knowledge-Conditioned Topology Generation for Multi-Agent Systems
• Learning from Hard Prompts: Difficulty-aware Advantage Amplification in Dynamic Sampling
• openJiuwen: Beyond Static Harnesses for Long-Horizon Coding Agents
• The Shape of Power: A Multilingual Framework for Social Power Reasoning in Dialogues
• SEPO: Evidence-Grounded Prompt Optimization via Structural Editing
• LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering
• MAIL: Memory-driven, Adaptive, Incremental, and Literature-grounded Framework for Hypothesis Generation in Chemistry
• Memristive-Friendly Hadamard Reservoir Computing: Structured, Multiplier-Free Recurrences at Scale
• Real-Valued Hyperdimensional Sequence Representations with Hadamard Product Binding and Shift Equivariance
• EvoUndo: Recoverability-Constrained Self-Evolution for LLM Agent Harnesses
• Propagating construction-time knowledge quality into medical question answering: A framework grounded in clinical guidelines
• Learning to Use Tools: Reinforcement Learning for Tool-Integrated Mathematical Reasoning
• RECAST: Recent & Context-Aware Sampling for Test-Time Adaptation in Streaming Biosignals
• Finding Where the Buck Stops: An Automated Failure Attribution-Based Reflection Framework for Multi-Agent Collaboration
• Regime-Aware Portfolio Management via Retrieval-Augmented LLM-Guided Expert Switching
• REINS: Refusal-Enhanced Inhibitory Steering with Sparse Autoencoder Features
• Automated Analysis Framework for Multilingual Climate-Health Literature Based on Multi-Agent Large Language Model
• GOD: Govern, Observe, and Direct - A Real-Time Control Room for Agent Societies
• AERA: Adaptive Evidence Residual Allocation for Efficient Test-Time Reasoning
• HyQuant: Hybrid-Precision Quantization for LLM Attention
• Credo: Reusable Declarative Primitives for Agentic Workflows
• Probing Perceptual Priors of MLLMs via Gibbs Sampling with Interpretable Generative Controls
• PCFBench: A Diagnostic Benchmark for Product Carbon Footprint Estimation
• MAP: A Benchmark on Multimodal Accessibility Planning for Real World Places
• GRACE:Gradient-guided Coreset Selection for LLM Unlearning
• AGENT-O: A Semantic Agent Card Framework for Interoperable and Governed Healthcare AI Agents