Recent research advances AI reliability, efficiency, and capabilities across diverse domains. In reasoning and agents, methods like PatchScout, NURESA, and OmniPiano improve data collection, memory transfer, and robotics, while CalVLA shapes policy confidence. Reliability tools include OmniMechanism for human-AI collaboration, Exo for self-improvement (reducing costs by 12-82%), and TAILOR for robustness certification. Efficiency gains are seen in XLA:TPU bug resolution, FastAgent CFD acceleration (1.53-15.36x), and State-Aware Activation Recomputation (18.16% speedup). Evaluation frameworks like LM6-CastBench, AgenticFinBench, and KierstonBench reveal gaps in geometry prediction, hallucination auditing, and crime investigation.
Agentic systems advance via procedural memory (Zep Agent Skills boosts pass rates by 4.5–18.8 points; I-PTC reduces token consumption) and adaptive workflows (SE2 achieves 26% higher accuracy; ChatDify reaches 0.96 success rates). Safety mechanisms include BioAgent-DF (recovers 39 decisions in cardiac assays) and Route-Guided CBF-QP Repair (enhances VLA safety). Limitations noted: Chain-of-thought oversight shows traces don't guarantee faithfulness; MemUse finds most memory entries irrelevant. Decentralized AI (OIM uses earnings-as-collateral) and service agents (NEIGHBOR raises expert scores to 7.74) show promise. Experimental design gains include F-CTSD (reduces violations by 4.95%). Verification tools like LARK verify 10/12 Level-1 kernels. RL methods (RLiD, SGR-OPSD) improve success rates on ALFWorld/WebShop. Safety studies reveal weak confidence-correctness separation (165) and silent PHI access in three models. Efficiency gains: ROCm InferaSim reduces config search to 9.1 GPU-hours; BasisKV offers 1.42x speedup. Scientific AI: OpenPerov achieves 27.05% efficiency; RE-SciVQA reaches 73.9% F1. Evaluation challenges: Four Bridges finds deception in 73% of multi-agent runs; CAREFUL-Bench shows 322/734 turns are destructive. Domain-specific advances: Grid Agents reduce costs by 33-67%; ScienceCUA boosts task success by 17.12 points; TriPhysBench shows embodied agents struggle at 50.38% accuracy on combined tasks.
LLM efficiency and reliability improve via CoGSeq-LTV for multi-business LTV prediction; Uni-LaSpec accelerates VLMs via load-adaptive speculative decoding (5.11x throughput); LHM improves long-to-short reasoning by 4% accuracy and reduces output length by 93.1%. Agent automation and evaluation show RSI-Dojo improves Qwen models by 8.74–16.87 PostTrainBench points; RAGLog localizes RAG faults with 0.73 pooled F1; EMBERS finds robustness issues 3.21x better than baselines; Simjecture solves the Z-pinch energy problem in 82.6 hours. Security and alignment studies reveal LLMs underrepresent malice, fake citations rise in post-trained models, and MCPGuard detects SDK conformance gaps. Multi-agent systems benefit from the Agent Board (reducing wait costs), ATLAS (improving memory accuracy by 5.7 points), and ATC (enforcing role contracts). Domain-specific applications include GlucoAgent (RMSE 22.5 mg/dL for glucose forecasting), OSGR (boosting Earth observation accuracy to 63.7%), and FP4Quant (1.56x throughput gain). Limitations noted include ATLAS's reliance on representation-level access and CONFIRM's focus on pre-specified plans. ValQID identifies web device vendors with up to 8.58x speedup and 47.1% macro-F1 gain. Overall, findings emphasize the need for rigorous auditing, efficient resource use, and robust evaluation to address complex real-world challenges.
Key Takeaways
- Exo self-improvement reduces costs by 12-82%.
- FastAgent CFD acceleration reaches 15.36x speedup.
- Zep Agent Skills boosts pass rates by 4.5–18.8 points.
- BioAgent-DF recovers 39 decisions in cardiac assays.
- Uni-LaSpec accelerates VLMs via 5.11x throughput.
- Simjecture solves Z-pinch energy problem in 82.6 hours.
- ATLAS improves multi-agent memory accuracy by 5.7 points.
- OSGR boosts Earth observation accuracy to 63.7%.
- Four Bridges finds deception in 73% of multi-agent runs.
- LHM reduces output length by 93.1% while improving reasoning.
Sources
- From Natural Language to Exact Cover: A Neuro-Symbolic Approach to Zebra Puzzles
- Reimagining Text-to-SQL: A Survey of Large Language Model Architectures and Strategies
- Early Stopping in Large Reasoning Models: A Survey and Empirical Audit of Efficiency and Reliability
- PatchScout: Thematic Web Data Collection via Information Foraging
- Neurons Are All You Need
- OmniPiano: Diverse Dexterous Piano-Playing Challenges for Standard, Robust, Safe, and Multi-Agent RL
- From Cross-Section to Cross-Omics: Conditional Cell Nucleus Image Editing in Spatial Transcriptomics
- MemU-Bench: Benchmarking GUI Agents on Long-Horizon User Interactions
- Beyond Semantic IDs: Atomic Creator Tokens for Large-Scale Generative Recommendation
- Quantifying Local and Global Fidelity of Nonlinear Embeddings
- Automated Theorem Generation in First-Order Logic by Construction from Triangular Standard Contradictions
- CalVLA: Shaping Native Confidence in Vision-Language-Action Policies with Outcome Supervision
- OmniMechanism Design for Human-AI Collaboration with Impressionable Minds
- Exo: An Architecture for Self-Improving Systems at Runtime
- Quantiles, Not Points: Streaming Quantile Matching for Ordinal-Aware LLM-as-a-Judge Calibration
- Robustness Starts with the Objective: Designing and Certifying Robust LLM-Generated Heuristics
- Few Trials, Many Metrics: Beta Mixtures for Low-Budget AI Evaluation
- Governing Equations of Human Sleep Macro-Structure:Symbolic Discovery from 1,106 Nights
- Low-Margin Branch Points in Deterministic LLM Deliberatio
- Skill++: Towards Lifelong Agents via Routing-Aware Skill Refinement
- State-Aware Activation Recomputation with Replay Contracts and Selective Preservation
- WAR-CRAFT: Refining Web API Recommendations through Requirement Auditing
- Autonomous Analysis of Thermoelectric Materials Big Data by Multi-Agentic AI and Construction of a Knowledge Base and Specialized AI
- Conversational Interaction for LLM-Based Clinical Text-to-SQL: A User Study with Healthcare Professionals
- XLA from the Trenches: Correctness on TPU Workloads at Scale
- LM6-CastBench: A Geometry-Aware Simulation Dataset for Transient Thermal Prediction in LM6 Sand Casting
- Designing Behavioral Profiles for Language Model Simulation of Out-of-Distribution Human Behavior
- FastAgent: A Physics-Aware FastFluent-to-Fluent Workflow for Rapid CFD Screening and Industrial Steam Dewaxing Design
- TopoQuant: Topological Framework to Characterize Behavioral Shifts in Quantized Large Language Models
- Designing Informative Actions: Action Construction for Long-Horizon Research Agents
- Pulse: Long-Horizon Tasks Need Streaming Personalization
- Inherited or Learned? Backdoor Selectivity in Distilled Language Models
- Reading Between the Chapters: A Graph-Free RAG Rebuilt With AI and Human Feedback
- AgenticFinBench: Benchmarking Financial Agents on Diverse Tasks from Real-World Requests
- Where Is the Truth When AI Agents Disagree? Dependence-Aware Belief Propagation for Unsupervised Truth Discovery (Extended Version)
- Kierston & KierstonBench: Simulation of Allentown, PA and Evaluation of Language Models in Crime Investigation and Surveillance Tasks
- HTCE: Homeomorphism-inspired Topological Cross-Scene Experience Canonicalization and Reuse for Object Navigation
- Reconstructing Fraudulent Strategies in Online Payments via Interpretable Inverse Reinforcement Learning
- The Pen is Mightier than the Pixel: Natural Textual Unlearnable Example for Large Vision Language Models
- VagueBench: Benchmarking the Robustness of Embodied Agents on Ambient Context Illusions
- Reasoning is topologically simple
- FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel Compensation
- Modelling lock-in of AI adoption through skill erosion
- ScamWorld: An Interactive Environment for Evaluating Web Agents Against Online Fraud
- Amortized Nonlinear Variable Selection with Prior-Data Fitted Networks
- The Platonic Brain: Representational Convergence in EEG Foundation Models
- LatentTriad: Pre-Action Risk Representations for Malicious Agent Skill Detection
- ALBI: LLM-Guided Inductive Logic Programming via Iterative Generate-Test-Constrain Refineme
- Creating Power Flexibility in Datacenters via Query Routing
- From Opinions to Commitments: Unstable Assignments and Silent Choices in AI-Mediated Proposals
- You don't need long context
- The Geometry of Ignorance: Detecting Suppressed Knowledge in LLMs through Activation Fragmentation
- JHiMA-WebRL: A Joint Hierarchical Multi-Agent Reinforcement Learning Framework for Long-Horizon Web Tasks
- The Evolution Strategies Paradox: Better Models from Worse Perturbations
- Structured Clinical Memory as Context for Probabilistic Decision Models: A Synthetic Evaluation with Jev
- A Guideline-Grounded Multi-Agent Framework for Personalized Exercise Prescription from Clinical and Wearable Data
- We Ran a Research Sprint With AI. The AI Worked. The Research Didn't
- Predictive gains, causal losses: why interventional competence cannot be learned from data alone
- LOLPlay: Benchmarking Real-Time Visual Agents in League of Legends
- Alignment Faking Depends on the Value, Not Just the Model
- Adversarial Speculative Reasoning Improves Long Horizon Agents
- AgentAM: An AI Agent for Spatter- and Defect-Aware Process Optimization in Laser Powder Bed Fusion
- Learning Behaviorally Relevant Dynamics in Multimodal Neural Recordings
- Catalyzing Interactive HTML Generation Capabilities via High-Fidelity Reinforcement Learning
- Meta$^2$: Precomputing Reusable Knowledge for Online Harness Evolution in Long-Horizon Agents
- Feeling of Knowing in Large Language Models
- Think Less, Locally: Local Preference Learning for Reasoning Compression
- RAHC: Rule-Augmented Hyper-Relational Knowledge Graph Completion for Materials Synthesis
- Reducing Logprob Without Causal Verification: Diagnosing SAE-Based PII Suppression in LLMs
- Spatial Memory for AR Glasses: Hierarchical Visual Acquisition and Source-Linked Recall
- Goal-Driven Autonomous Optical Experimentation via a Self-Recursive Hardware-Software Harness
- Treating the Symptom, Missing the Cause: Anatomy of a Scientific Reasoning Failure
- Bidirectional Centered Score Matching for Discrete Diffusion Distillation
- Harness-S: Learning from Sibling Harnesses for Generalizable Agent Training
- AceAction-5M: Hierarchical Goals for Generalist Action Learning Across 3D Game Worlds
- LaplaceDiag: Fault Diagnosis through Changes in Learned Actuator Effects
- El Agente F\'armaco: A Scalable AI Scientist for Drug Discovery
- Unsatz: Evaluating Scientific Discovery Without a Given Ansatz
- DeliverBench: Benchmarking AI Agents on Delivering Realistic Multi-Stage Professional Projects
- LAND: Learning an Adaptive Number of Draws for Bayesian Best-of-N Selection
- BLPO: Beyond-Local Policy Optimization for Long-Horizon LLM Agents
- Why Tabular Foundation Models Excel on Relational Data They Have Never Seen
- User Attributes and Political Ideology Shape LLM Receptiveness in Online Political Debates
- Beyond Answers, Traces, and Efficiency: Evidence for Adaptive Reasoning
- Human-Oriented World Model
- The SIMPLE Law for Energy Cost of LLM Inference
- Networked AI: Architecture, Theory, and Applications
- Chain of Faith: Policy-Internal Rewards for Reasoning Reliance
- Appointed-Time Fault-Tolerant Fast Terminal Sliding Control for UAVs with a Tracking Error-Based Disturbance Observer
- AdKnob: Ad Intensity Control and Labeling for LLM-Native Advertising
- A Discrete-Time Variational Objective for Degenerate Diffusion and Equivalence to Flow Matching
- BridgeDock: Taming Generative Co-Folding Prior for Molecular Docking with Diffusion Bridge
- A Tool in the Forward Pass: Frozen Tape Modules as Layers of a Small Language Model
- Decision-Centric Research: A Calibrated Decision Model and Benchmark for Scientific Agents
- Trail: Accurate Local Smart Home Control Using Decision Models
- Editing the Goal: Per-Token SAE Steering and Its Depth Limits in $\pi_{0.5}$
- X-World 1.6: Building Controllable Virtual Cities with World Models
- Evolving Scientific Intelligence: From Fragmented Evidence to Distributed Experimental Loops
- Nonconformal Distortion and Local Orientation Reversal in the Fourier Circuit of a Modular-Addition Transformer
- Hacking or Just Failing? A More Capable LLM Monitor Better Separates Reward Hacking from Honest Failure
- Dependence-Aware Risk Assessment and Optimal Design of AI-Agent Safeguard Systems
- Efficient NPN Classification for Large Functions using Graph Neural Network
- Decentralized Multi-Agent Coordination and Communication in Shared Codebases
- A generative foundation model for computational brain imaging
- Learning to bid and bundle: a posterior-sampling agent for online retail
- A Generalized Sparse Fuzzy Clustering Framework with Automated Cluster Cardinality Estimation Ability
- From AutoResearch to Continual AutoResearch: Do Agents Learn Across Tasks?
- Artifact-to-Capability: Diagnosing Breaks in Capability Formation of Self-Evolving Agents in a Controlled Environment
- Signals: Continuous Diagnostics of AI Agents
- Silent Instrument Errors Reorder Agent Leaderboards
- Human-to-Robot Transfer via Bridging Representation and Action Space in VLAs
- Benchmarking Unsupervised Jamming Detection in 5G Open RAN
- Decoupled Predictive Decision Making: A Shared Conditional Predictor for Agents
- State-Aware Virtual Patient Simulation for Training Psychiatrists in Challenging Clinical Scenarios
- Agentic Context Management for Multi-Turn Human-AI Conversations
- To Jev or Not to Jev: When Can System-1 Decision Models Replace LLM Judges?
- Swap, Replicate, Read the Mass: A Protocol for Auditing Demographic Bias in System One Decision Models
- Compression Technique Interactions in Small Language Models: A Factorial Ablation Study
- One Offset, Two Directions: Baseline Adjustment via Subgroup Offsets
- Who Schedules Agent Memory? A Sidecar Approach to Long-Horizon Tasks
- Leap-VLA: High-Rate Robot Control from a Low-Rate VLM Backbone
- GURU: A Hierarchical Neuro-Symbolic Multi-Agent System for Verification-Aware Automation of Mission-Critical Digital Engineering and Operations
- VCoT-SQL: Enhancing Text-to-SQL Generation through Dual Chain-of-Thought Validation Training
- AI Data Analysis: A New Approach - Multi-engine Joint State Data Analysis
- Benchmarking Machine-Learning Classifiers for Neoantigen Immunogenicity Prediction and the Limits of SHAP Interpretability
- Can Agentic AI Undermine Its Own Assurance Loop? Closed-Loop Dependency Assurance for Cognitive Network Digital Twins
- Temperature, Not Knowledge: What Recursive Self-Training and Its Remedies Do to a Language Model's Confidence
- Constraint Dependency Graph Retrieval for Compliance Detection on Long SKILL Documents
- IJev: Order-Invariant Decisions via Iterated Cyclic Reordering
- EMG-Hybrid: A Dual-Engine Elite Memory--Guided Adaptive Optimization Framework for High-Dimensional Feature Selection
- Compile as Hypothesis: A Headroom Typology for Prompt Optimization on Real Reasoning Models
- HiGenTrans: Hierarchical LLM Knowledge Transfer for Generative Recommendation
- Robust Reward Machine Learning and Execution in Noisy Environments
- Concept-Guided Prompting Strategies Increase Diversity in LLM-Generated Recipe Populations
- Urban functional knowledge transfer through global representations amid uneven data
- FOCUS: Closed-Loop Attention Feedback for Efficient Vision-Language Understanding
- Learning by Doing: Autonomously Compiling and Curating Agent Skills from Verified Traces
- A Self-Exploring and Self-Evolving Paradigm for Adaptive Edge AI Development
- \textsc{ChatDify}: Automating End-to-End Dify Workflow Synthesis from Natural Language via Multi-Agent Collaboration
- From Tool State to Scientific Action: Auditing Decision Provenance in Biomedical Research Agents
- I-PTC: Interactive Programmatic Tool Calling for Stateful Tool-Augmented Agents
- SPAR: Experience-Driven Strategy Evolution in Multi-Session B2B Sales
- Xiaomi Auto World Model 2.0: Causal World Models for Real-World Driving and Embodied Intelligence
- Local or Escalate? Post-Training Expert Routing for Real-Time Duplex Speech Models
- Chain-of-Thought Oversight Requires Separate Evidence for Faithfulness and Monitorability
- Delivered, Not Used: Evaluating Agent Memory at Its Point of Use
- Route-Guided CBF-QP Repair for Safe Vision-Language-Action Models
- Just-in-Time Skill Retrieval via Long-Horizon State Tracking
- A Compute Center Anyone Can Join: A Peer-to-Peer System for AI Inference
- Self-Evolving Agentic Workflows via Semantic Decomposition of Complex Tasks
- NEIGHBOR: Aligning Service Behaviour and Evidence Use in an LLM Agent for Community Resident Services
- Fairness-Aware Design for Contextual Experiments: Guaranteeing Reliability and Equity in Heterogeneous Subgroups
- Geometric Morphology Learning: Empowering Foundation Models from Structure to Spacetime Intelligence
- Who Asks, Who Answers: Linear Attention Builds the Query in Hybrid Language Models
- Opening the Black Box of Neural Auction Mechanisms
- ClickSight: An LLM Framework for Interpreting Student Clickstreams Through Learning Strategies
- To Align Long, Plan Short: Simulating Diverse Profile-Aligned Students by Realizing Short-Term Plans in Long-Horizon Learning Environments
- Safety4DM: A Vision-Language Framework for Risk-Aware Dynamic Maps in Vehicle-Road-Cloud Collaborative Driving
- SWE-Trail: History-Guided Trajectory Collection for Coding Agent Training
- LARK: Lean-Assisted Reasoning for LLM-Generated GPU Kernels
- RecGroupMemory: Feedback-Aware Cohort Memory for E-Commerce Recommendation
- Inner-Speech Decoding from EEG Across Three Corpora Under a Fixed Architecture and Protocol
- Rewarded for Not Getting Caught: Why Alignment Training Selects for Conditional Compliance and Cannot Instill Norms
- ParetoSaddler: Pareto-Guided Optimization of Agent Harnesses for Quality and Cost
- The 3-Tier Integrated Technology Risk Model for Occupational Safety and Health
- Calibration-Constrained Diagnostics for Evaluating LLM Self-Evaluation
- TIGER: Learning from Researcher Taste to Generate Tailored Research Ideas
- RLiD: Reinforcement Learning in the Decision Plane for Language Agents and LLMs
- Accessed, never disclosed: Silent access to protected health information by medical artificial intelligence agents
- SGR-OPSD: Learning from Subgoal Plans with Graph-Regularized Credit Assignment for Agentic Reinforcement Learning
- Towards Multi-Value Alignment via Value Localization in LLMs
- Benchmark of THReD: Evaluating an LLM Agent and Its Preprocessing Tools for Turning Electronic Health Records into AI-Ready Datasets
- Think Globally, Generate Personally: Multi-View Thinking with Adaptive Generation for Generative Recommendation
- FORGE: Fault-Oriented Rework Code Generation for Executable Interlocking Assembly
- A balloon effect: ban one channel and a hidden trait finds another
- Controlling Action-Head Co-Adaptation to VLM Representation Anisotropy
- WASP-Mem: Let Agent Organize Procedural Memory. Human Spend Effort on Diagnosis
- Thinking Fast, Trading Late: System One Decision Models vs the Market
- ROCm InferaSim: Cut LLM Serving Trial Costs and Deploy the Best Configuration
- When Prediction Shapes the Future: A BlueSky Agenda for Reflexive Time Series Analysis
- Parallelism-Aligned Pruning and Round-Aware Context Reuse for On-Device Multimodal Agents
- BasisKV: Value Latents for Communication and Key Routing
- Twice as Fast, Until It Isn't: Quantized Serving of Mid-Size Open LLMs
- Embodied AI for Scientific Discovery: From Physical Action to Scientific Evidence
- Variable names provide priors: semantic grounding enables data-efficient numerical prediction by large language models
- Parametric CAD Bench v3: Evaluating Agentic Parametric CAD Authoring from Text and Engineering Drawings
- SearchInject: Retrieval Exposure Control for Indirect Prompt Injection in Language-Model Search
- GeoGuesser Bench: Benchmarking Vision-Language Agents on Interactive Geolocation
- Reap: Uncovering Monte Carlo proof search with a lightweight model
- Auditing Representation-Space Fraud Detection: A Bayesian IRL Framework for Calibrated Uncertainty and Its Limits
- UPSHIFT: User Preference Steering from Human-AI Interaction Feedback Traces
- Is your LLM an Agressive Narcissist with Depression? Analyzing Psychological Portraits of Large Language Models
- AffToM: Benchmarking Recursive Affective Theory of Mind in Dynamic Multi-Party Interactions
- MURFormer: Attention-Based Lightweight Multi-Scale Uncertainty-Guided Refinement Framework for PPG Denoising
- Human-Supervised Artificial Intelligence Researchers in Electric Power Systems: Knowledge, Physics, Simulation, and Governance for Large Language Models
- Analysing the Narrative Paradox through Formal Models and Properties in Interactive Narrative
- DrawOffer: Evidence-Grounded Longitudinal User-State Modeling for Career Support
- Offline Approximate Value Iteration: A Reinforcement Learning Approach for Satellite Constellation Deployment
- CARE: Grounded Recursive Self-Improvement for Production Customer-Service Agents
- Think with Skills: Pseudocode Planning Before Execution for Advanced Reasoners
- Value-Conditioned Proposal Acquisition for Test-Time Reasoning
- SpecOPD: Branch-Parallel On-Policy Distillation for Speculative Decoding
- JABER: Sequential Quality Control for Human-AI Data Annotation under Shared Bias
- What makes a theorem interesting? Operationalizing interestingness for open-ended exploration
- CAREFUL-Bench: Effect-Aware, Judge-Free Evaluation of Multi-Turn LLM Function Calling
- Action Verification through Conjecture and Refutation for Language Model Agents
- PRIME: Predictive Routing of Interventions by Modeling Effects for Scientific Ideation
- A Fine-Grained Hierarchy and Benchmark for State Tracking in Language Models
- Avatar Intelligence: Amplifying Human Agency Through Persistent Personal Agents
- Towards a Trustworthy and Governable Internet of Agentics: An Architectural Perspective on Protocols, Systems, and Services
- CoworkOS:Scaling Agent Coworkers for Real-world Automation
- CUA-Master: Scaling Computer-use Agents through Open-World Recursive Improvements
- ComplianceAtlas: A Taxonomy-Driven Agentic AI Framework for Automated AI Act Gap Analysis and Remediation
- From Local Skills to Long-Horizon Web Tasks: Progressive Skill Exploration for LLM Web Agents
- InterFuse: Interaction-Guided Adaptive Fusion of Heterogeneous Models for Payment Risk Ranking
- Axion 1.0: Open Agent Foundation Models, Simplified
- From FAIR to AI-FAIR Metadata-First Infrastructure for AI-Enabled Science
- Top-$k$ Retrieval Leaves Memory Value Undetermined
- ZeroGraph: Compositional Blockwise Training via Materialized Boundary Supervision
- Mixed-Methods Exploration of Sociotechnical Factors to Improve Adoption of Crime-Reporting Apps in Urban Communities
- Can We Stop the Backtest Yet? Anytime-Valid Promotion Tests for Adaptive Financial Research Agents
- Same Outputs, Different Scores: Auditing the Unchecked Graders of Multimodal LLM Benchmarks
- Grading the Unannotated: Insights from the WorkRB Skill Extraction Challenge at RecSys-HR 2026
- Online Label-Free Gradient Matching for Continual Language Model Training
- Playing Dumb: How a Chess Transformer Simulates Users Across Skill Levels
- Optimize the Harness, Not the Model: Fitting a Minimal Agent to a Self-Hostable LLM for Investigative Analysis
- From Small to Large: Assessing Medical Reasoning, Benchmark Contamination, and Model Size in Language Models
- HybridSpec: Throughput-Aware Speculative Decoding for Hybrid GDN-MoE Models
- Can Jev Improve Demonstration Retrieval for In-Context Learning?
- Evidence-grounded large language models guide perovskite solar cell design and optimization
- Taurus: LLM-in-the-Loop CT Annotation Matches Coding Agents at Half the Cost
- DGC-Net: Cross-Modal Dynamic Routing with Global-Guided Compensation in Multimodal Sentiment Analysis
- TactiNavi: Serving and Revising an LLM Game Coach with Explicit Strategy Conditions
- Are LLM Assistants Regulatory Ready? Answers from regenold's 2026 EU AI Act Q&A Challenge
- AuroraWorld: Structured 4D World Modeling for Exo-to-Ego Video Generation in Robot Manipulation
- A research world model of power engineering learns its dynamics and reveals what-ifs
- RE-SciVQA: Improving Scientific VQA With Context-Relevant Retrieval and Lightweight MLLM Adaptation
- What Do LLM Judges Actually Verify? Matched Controls and Record Gaps for Document Currency
- Strategist: Evolving Reasoning with Agents
- DiabAgent: Toward Agentic Physiological Control
- Four Bridges: Measuring Deception Propensity in a Multi-Agent Game
- Predictive Safety Guardrails Do Not Always Make LLM Agents Safer
- Creating Hybrid Interactive Narrative Systems with LLM: From Formal Description to Evaluation
- SelfImprove-Bench: Evaluating Generalization in Self-Improving Agents
- Stage-aware Multimodal Brain Graph Contrastive Learning for Neurodegenerative Disorder Diagnosis
- Measuring AI Agents as Teammates with Collaborative Intelligence
- Widening Harm Evaluation Coverage with a Judge that Reads the World
- Probing Local Conditional Fit for Diffusion Model Membership Inference
- EEFA: A Framework for the Ethical Evaluation of AI Agents
- Cumulative and Audited Guardrail Tool Evolution for LLM Agent Safety
- LEANGUARD: Lean-Verified Guardrails for Tool-Using Agents
- Toward Agentic AI-Ready Workflows in Built Environment Organizations: Knowledge Structures and Interoperability Challenges
- Data-Scarce, Multi-Granular, Multimodal: Benchmark Design for Biosynthetic Gene Clusters
- Beyond Binary Detection: Type-Aware Hallucination Analysis and Hierarchical Detection in Financial Question Answering
- Recursive Reasoning Exploration: Mathematical Research through Evolving Reasoning Graphs : case study: An Exact Certificate-Degree Theorem for Two-Term Tensor Decompositions : Phase III of the Math Instruments Program
- RubricWorld: World Model-Guided Rubric Optimization for LLM Graders
- Hippocampal consolidation without replay in fully fine-tuned networks
- Beyond Multimodal Emotion Conflict: Temporal Context-Aware Hidden Emotion Detection with Multimodal Large Language Models
- TriPhysBench: Evaluating Physical Intelligence for Embodied Agents
- Is Your Long-Horizon Benchmark Really Deep? Measuring and Synthesizing Depth in Terminal Tasks
- Does reflective LLM feedback survive a wet-lab-scale label budget?
- WebHarbor: Docking Real Websites for Evolving GUI Agent Environments
- FedReady: Bounded Agents for Federation Formation
- An Agentic Framework for Automated Power System Studies and Grid Connection Assessment
- Can AI Do Formal Demography? Why Vaccinating the Oldest Saves Both the Most Lives and the Most Years of Life
- ScienceCUA: Resource-to-Skill Construction and Execution-Guided Skill Evolution for Scientific Computer Use Agents
- Auto-Research Agent on MSA Organization to Improve Protein-Protein Complex Structure Prediction
- Toward Knowledge-Centric AI: Discovering, Internalizing, and Using Explicit Knowledge Across Agents and Generations
- LUCA: A Synthetic World for Experimental Reasoning
- Bridging Low-Level Image Fusion and High-Level Semantic Understanding
- Integrating Domain Adaptation and Retrieval-Augmented Generation for Saline-Alkali Land Reclamation Planning
- TLAPS-Bench: Formally Proving Correctness of Real-World System Specifications with TLA+ Proof System
- RxnMiner: A Multimodal Agent for Structured Reaction Data Extraction from Chemistry Figures
- AttentionIndex: Accelerating Learned Sparse Attention with a Causal Inverted-File Index
- MAGAR: Multimodal Agents for GUI-Based 3D Mechanical Assembly and Minimal Rework
- CoGSeq-LTV: Scalable Missing-Aware Graph-Sequence Modeling for Multi-Business Transaction Analytics
- MemHarmBench: Evaluating Cross-Session Prompt Injection in Memory-Persistent AI Agents
- Large language models underrepresent malice and overreport virtue
- Are Fake Citations a Form of Reward Hacking? A Pre-Registered Study of Post-Training and Citation Cues
- OntoDoc: Automatically Generating Ontology Resource Descriptions
- Capacity Planning for Resource-Constrained Clinical LLM Serving
- CivicBench: A Benchmark and Memory Framework for Web-Based Civic Data Analytics
- Seeing What Matters, Verifying What Pays: Load-Adaptive Speculative Decoding for Vision-Language Models
- CARD: A Proof-Carrying Certificate-to-Action Interface for Relative-Measurement Graphs
- Functional-Group Knowledge Is Structured in Language Models Before It Is Spoken
- MCPGuard: Finding Conformance Gaps in MCP SDKs via Dual-Track Analysis and Dynamic Adjudication
- Training LLM Agents to Automate LLM Post-Training with RSI-Dojo
- From Logs to Faults: Reference-Free Fault Localization for Retrieval-Augmented Generation Pipelines
- EMBERS: Empirically Motivated Robustness Testing of Deep Research Systems via Defect-Guided Fuzzing
- Counterexample-driven autonomous research with application to excess radiated energy in a Z-pinch
- Performance, Scalability and Maintenance: An Experimental Study of RPA vs. Autonomous Agents in the Healthcare Domain
- Low-Rank Hierarchical Merging for Efficient Long-to-Short Reasoning
- Measuring Model-Specific Alignment Drift in LLM-Based Opinion Dynamics
- What Is This Memory Useful For? Measuring Task-Conditioned Utility in Long-Term Agent Memory
- ProofPaths: An AI Agent Framework for Assessing Student Reasoning
- Exploit or conform? How norm utility prompting changes the way LLMs use community beliefs
- Recursa: Toward Recursively Self-Improving Foundation Model Systems
- Look Before You Launch: Learned Throughput Prediction for Hardware Selection in a Production Training Fleet
- Forgetting to Improve: Principled Data Removal in Active Learning
- On the Primacy Bias in RLVR Training
- Query-based Submodular Data Selection for Efficient Instruction Tuning
- The Agent Board: Kanban for Machines; Durable Coordination for Agents That Outlive the Conversation
- AlgoPilot: Cross-Paradigm Reasoning in Language Models via Strategy Selection and Guidance
- Validation-Gated ML Pipeline Synthesis with Expert-Grounded Self-Evolving Skills
- Interpretable Early Termination of Web Navigation Agents via Closed Sequential Pattern Mining
- GlucoAgent: Personalized Glucose Forecasting from Multimodal Personal Context
- Toward User-Centered Agentic Recommendation: When Does Platform Competition Benefit Users?
- Exact Optimality Certification of QPLIB 2096 through Complete Set Partitioning
- Reliability Stress Testing of RAG Retrieval under Query-Relevant Corpus Poisoning
- Neologism Inoculation Prompting
- ATC: Agent Trajectory Control for Role-Contract Enforcement in Black-Box Multi-Agent Systems
- Drift No More: Spectral Locking for Long-Horizon Open-Weight Multi-Agent AI
- OSGR: Observation-Support Graph Reasoning for Earth Observation Question Answering
- FP4Quant: Efficient 4-Bit Communication Compression for Model-Parallel LLM Training
- Beyond Fine-Tuning: Evaluating Prompt Engineering and Prompt Tuning for Roman Urdu Hate Speech Detection
- Patching the Protocol: A Taxonomy of Pre-Registration Repairs in LLM-Proposer Experiments
- CORA: Teacher-Guided Causal O-RAN Adaptation for Jamming Detection
- CaliSplat: Sparse Gaussian Fitting and Calibrated Rate Allocation for Image Compression
- Correct Labels Are Not Enough: Reliable Metadata Can Silently Control Model Behavior
- CogniLife: Continuous Cognitive Distillation for Deployed Multi-Expert Agent Marketplaces
- A Parametric Linear Conic Projection for Structured Sparse Attention
- Large Language Model-Assisted Adaptive Heuristic Search for Agile Earth Observation Satellite Scheduling
- CONFIRM: Governing Scientific Reporting for Trustworthy Agentic Neuroimaging
- DualWorldBench: Can Agents Plan Deliveries across Symbolic and Grounded Worlds?
- DoGNAVY-Exploit: Harnessing Parallel Open-Weight Agents for Stage-Aware Vulnerability Exploitation
- Clocks all the way down: does sequence-processing AI need endogenous temporal structure?
- Palimpsestic Accountability in the US Federal AI Inventories
- Reconstruction-Based Diagnosis through G\"odel-Regularized Virtual Observability
- Formalizing PTX in Lean to Prove GPU Kernel Correctness
- Verifier strength does not buy discovery quality: a pre-registered two-domain test of LLM-based automated research, with post-hoc evidence pointing to proposer-pool composition
- Fast, Cheap, and Fair? Evaluating Decision Models
- Learnings and Pitfalls from AI-Assisted A/B Experimentation at Scale
- Inter-Agent Influence: Evaluating Persuasion, Deception, and Coercion in Multi-Agent Systems
- Planning-Coding Separation in LLM-Based Code Generation: An Empirical Study
- When Should an LLM Steering Edit Depend on the Input?
- SAT Encoding for Set of At-Most-K Cardinality Constraints with Ladder Shape
- ValQID: Learning the Net Value of Semantic Query Groups for Web Device Identification
Comments
Please log in to post a comment.