AI News

Multi-Token Prediction and Alternative Tokenization Improve Learning Latent Multi-Hop Graph Search at Scale (arxiv.org)

Drift Says When, Budget Says How Much: Rate-Controlled KV Re-selection for Long-Form Multimodal Generation (arxiv.org)

Predictable performance is suboptimal, and what it means for model routing (arxiv.org)

MOER: Mix-Policy Optimization under Expert Prefix Guidance with Selective Rollback (arxiv.org)

SLoSa: Small, Low-Power, and Safe Foundation Models for Robotic Task Planning (arxiv.org)

StyleStream 2.0: Fast and Controllable Streaming Voice Style Conversion (arxiv.org)

SANE: Automated Scorer-Specific Metamorphic Testing for LLM Judges (arxiv.org)

inr2action: Encoding Robot Episodes as Neural Implicit Representations for Recall and Action (arxiv.org)

Metric Conformal Prediction via the Conditional Expected Local Radius (arxiv.org)

A Small Model Asks Better: Fine-Tuning Query Clarification for Retrieval-Augmented Generation (arxiv.org)

Solving Go under Positional Superko: Sound Proof Transfer and New Board Values (arxiv.org)

AtomicRL: Atomic Reinforcement Learning for Long-Horizon Robotic Manipulation (arxiv.org)

Parallel AI Agent Collaboration as Graph-Structured Distributed Block Optimization (arxiv.org)

Are EEG Sequences Time Series? EEG Classification with Time Series Models and Joint Subject Training (arxiv.org)

Evolutionary Deep Nets for Non-Intrusive Load Monitoring (arxiv.org)

On Sex, Evolution, and the Multiplicative Weights Update Algorithm (arxiv.org)

Sensorimotor Alignment: A Platonic Proxy for Embodied Policy Success (arxiv.org)

SpectralCanary: Embarrassingly Simple Low-Frequency Traces for T2I Dataset-Use Auditing (arxiv.org)

EpiSEM-Bench: Evaluating Safety and Evidence Recovery in LLM Agents under Missing Evidence (arxiv.org)

Beyond Best-of-$k$: Prior-Consistent DAPS for Ambiguous Nonlinear Inverse Problems (arxiv.org)

Would an LLM Match into a Residency? Uncertainty, Reasoning and Image Use of Open Language Models on the Spanish MIR Examination (arxiv.org)

LACToS: Local Anomaly Detection via Conditional Typicality of Superpixels (arxiv.org)

Direct Inference on Admissible Sets (Diona) (arxiv.org)

From Reasoning to Action: Benchmarking Frontier AI Models on LIBERO-10 Robot Manipulation Tasks (arxiv.org)

SWE-DIVE: Multi-Turn Coding Agent Evaluation with Dynamic User Simulation (arxiv.org)

SP-HLM: Sparse-Pose-Driven Motion Generation and Whole-Body Control for Humanoid Loco-Manipulation (arxiv.org)

What the Author Wrote, What the Agent Reads: Silent Cue Loss in Enterprise Document Ingestion (arxiv.org)

Stationarity Constraints Improve Both Energies and Densities of Machine-Learned Exchange-Correlation Functionals (arxiv.org)

Keeping Time vs. Understanding Time: Neither Alone Explains Temporal Reasoning in LLMs (arxiv.org)

Structure-preserving reduced-order models of thin panels in supersonic and hypersonic flow (arxiv.org)

MoG-VLM: Efficient Video Anomaly Detection through Motion-Guided Vision-Language Models (arxiv.org)

Prompt Sensitivity in Tool-Augmented Model Comparisons (arxiv.org)

GM-GReFEL: Geometry-Guided Spatiotemporal State-Space Learning for Robust Dynamic Facial Expression Recognition (arxiv.org)

RegoMender: A Multi-agent Framework for Automated Rego Policy Repair (arxiv.org)

Beyond Small Talk: Modeling Disclosure Trajectories in Human-AI Conversation (arxiv.org)

Enhancing Cross-Grade Generalization in Banana Variety Classification via Hybrid CNN-Transformers and Knowledge Self-Distillation (arxiv.org)

VeriEvolve: Evaluator-First Evolution for Tool-Using Agents (arxiv.org)

A Strong MoE Is All You Need for Process Supervision (arxiv.org)

Mixing Matters: Evidence Position Bias Across Sequence Mixers in Long-Context Question Answering (arxiv.org)

Post-Audit Mirage: Identifiability Limits of Offline Agent Verification (arxiv.org)

Delta N-gram: Forward-Only Continual Writing into a Language Model's Native N-gram Memory (arxiv.org)

DRAFT: Decoding and Reconstructing 3D Room Layouts from fMRI and Camera Trajectories (arxiv.org)

NN-Based Transistor Modeling for Ultra-Fast SPICE Simulation and Standard Cell Library Characterization (arxiv.org)

Mostly Reading: What Open-Book Scores Measure in Document-Level LoRA Adapters (arxiv.org)

Diagnostic Loss: Asymmetric Supervision for Radiology Report Generation (arxiv.org)

Piano2Pop: Expanding Piano Demos into Pop Productions Without Aligned Supervision (arxiv.org)

Training Small Language Models for Transferable Tool Selection (arxiv.org)

Evaluating the Applicability of LLM Inference Optimizations on Apple Silicon (arxiv.org)

Phase Transition from Random Features to Fully Trainable Networks (arxiv.org)

SympFormer: Inertial Attention Blocks from Damped Hamiltonian Dynamics on Spaces of Probability Measures (arxiv.org)