AI News

How Many Ranks Does a Token Need? A Noise-Floor Squelch for Rank-One LoRA Experts (arxiv.org)

Who Should We Listen to More? Welfare-Aware Preference Acquisition for Pluralistic Alignment (arxiv.org)

Lag: A Mean-Field Optimal Policy for Online Assignment with Reusable Resources (arxiv.org)

Graph Coarsening Enables Escaping Local Optima in Compute-Constrained Nonconvex MAX-CUT (arxiv.org)

Should SLMs be in the Driver's Seat? A Study of Local Models as Interpreters in Cloud-Written Agent Programs (arxiv.org)

Long-Horizon-Terminal-Bench 2.0: Automating Diverse Task Construction and Measuring Sustained Agent Progress (arxiv.org)

VETNET: Assembling Benchmark Datasets for AI-Driven Veterinary Innovation (arxiv.org)

Unconstrained CTI-REALM with Open-Weights (arxiv.org)

Visualizing the World Model in an Agent's Value Function (arxiv.org)

Deep Generative Models for Phylogenetic Inference with Complex Evolutionary Processes (arxiv.org)

How Inner Optimizers Shape Replica Scaling (arxiv.org)

Theoretical and Practical Advances in Hierarchical Bayesian Deep Learning Methodology (arxiv.org)

Cross-Modal Latent Influence: Patch-Level Training Sample Attribution for Vision-Language Models (arxiv.org)

Refined Topology-Pass Scheduling: Damped Trust-Region Compensation for Dynamic k-NN Graphs in SE(3)-Equivariant Structure Prediction (arxiv.org)

Does a Governed Confidence Score Know When a Text-to-SQL Answer Is Right? An Exploratory Study on BIRD and a Pre-registered Confirmation on Spider (arxiv.org)

Beyond Aggregate Accuracy: Item-Level Transitions in Prompt Optimization (arxiv.org)

TRACER: Trace-Regularized Training Against Indirect Prompt Injection (arxiv.org)

Contextual Credit Policy Optimization (arxiv.org)

Genesis Steering: Better Activation Steering Through Training Data (arxiv.org)

Which Metadata Matters for LLM Tool Selection in MCP? Evidence from Controlled Factor-Level Interventions (arxiv.org)

Harness^2: Recursive Agent Harnessing for an Open World (arxiv.org)

Echo: Error-Compensated 4-Bit Tensor-Parallel Communication for LLM Training and Inference (arxiv.org)

Controllable Diverse Sampling through Pullback Directions (arxiv.org)

ConicFlow: constrained Inference-time optimal control for flow matching models (arxiv.org)

Unlocking Structured Pruning for Reasoning Language Models via Attention-Preserving MLP Trimming (arxiv.org)

Actionable Meta-Requirements for AI Robustness Auditing (arxiv.org)

MA-Bench: Measuring Operational Planning in Language Models (arxiv.org)

Towards Training and Inference of Graph Neural Networks in Graph Databases (arxiv.org)

Demystifying On-Policy Distillation: Where Does the Teacher's Advantage Live? (arxiv.org)

Diagnosing Low-Confidence Patient Profiles via Localized Harmful Covariate Shift Detection (arxiv.org)

PairIT: Autoregressive Transformers for Low-Data Molecule Optimization (arxiv.org)

When Pleasing the User Compromises Care: AI Sycophancy in Caregiver-Mediated Healthcare (arxiv.org)

ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence (arxiv.org)

ANASSA: From High-Level Framework to Low-Level Architecture for Spatial Agentic AI (arxiv.org)

Variance-Optimal State Resampling for Reinforcement Learning with Verifiable Rewards (arxiv.org)

Measuring the Inference Energy Efficiency of Open-Weight LLMs across Benchmarks, Reasoning Modes, and Hardware (arxiv.org)

SAT-RL: Training Language Models for Logical Reasoning with Symbolic Solvers (arxiv.org)

Frustratingly Easy Agentic RAG via Portable Skills: A Case Study in Customer Support (arxiv.org)

Energy Ring: Stabilizing Feature Energy for Reliable Test-Time Discovery (arxiv.org)

AlignPoison: Conditional Targeted Poisoning of Generative Recommenders via Decoding-Aligned Prefixes (arxiv.org)

Adaptive Computation with Conformal Certification (arxiv.org)

Spectral-MoE: Heterogeneous Per-Layer Expert Utilization in Dense-to-MoE Conversion (arxiv.org)

LLMs program using Reusable and Composable Abstract Representations (arxiv.org)

Tracing Morphology, Semantics, and Divergence in Language Acquisition (arxiv.org)

MvFFN: Multi-view Floor plan Feed-forward Network for Unposed Wide-Baseline Panorama Layout Reconstruction (arxiv.org)

HiPPT: Physiology-Grounded Wearable PPG Representation Learning via Multi-Scope Text Alignment (arxiv.org)

Reels-RSI: Program-Level Self-Improvement for Code-Rendered Explainer Videos (arxiv.org)

MontageFlow: Physically Valid High Fidelity EEG Generation (arxiv.org)

ScienceBuddy-PV: A Verifier-Governed Recursive Improvement Protocol for Panvascular Intervention Research (arxiv.org)

FDE-Bench: Evaluating End-to-End Delivery from Underspecified Real-World Business Requests (arxiv.org)