AI News

AlignQuant: Tile-Aligned Mixed-Precision Quantization for Efficient LLM Generation (arxiv.org)

No Transformer Beats Six Covariates: Long-Horizon Prediction of Depressive Symptoms from Childhood Essays (arxiv.org)

Inference and learning in sparse autoencoders as natural gradient flow (arxiv.org)

WAMJET: A Harness for World Action Model Acceleration (arxiv.org)

When Forgetting is not Catastrophic: On the Mechanics of Spurious Forgetting (arxiv.org)

Co-Evolving Paths and Flows via Path-Flow Alignment (arxiv.org)

Secure Speculative Decoding for Large Language Models (arxiv.org)

FedDermaSeg: Federated Learning for Dermatological Image Segmentation (arxiv.org)

Towards the Automatic Synthesis of Interpretable Chess Tactics (arxiv.org)

WildMatch: Weakly Supervised Image Matcher Adaptation for Wildlife Re-Identification (arxiv.org)

When Can World Models Recover Physical Laws? (arxiv.org)

How Much Do LLM-as-a-Judge Design Choices Matter? A Systematic Comparison of Prompt Designs, Rating Scales, and Models (arxiv.org)

OncoNoteBERT: A Foundation Representation Model for Natural Language Processing of Real-World Outpatient Oncology Notes (arxiv.org)

Learning Explainable Representations of Complex Game-playing Strategies (arxiv.org)

Have I Seen Enough? Frozen Video-Language Models Encode Evidence Readiness (arxiv.org)

Latent space bias directions in LLMs capture confidence, not fairness (arxiv.org)

Systemization of Knowledge (SoK): Human-Centered AI Safety for Youth (arxiv.org)

AnyBottle: A Recipe to Only Keep the Concepts You Really Need (arxiv.org)

DeltaTTT: Layerwise Optimization for Nonlinear Recurrent Memory (arxiv.org)

Sherpa: Teaching LLMs to Teach Adaptively (arxiv.org)

Agent in a Bottle: Can LLM Agents Turn Their Capabilities Into Cheap, Scalable Artifacts? (arxiv.org)

VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning (arxiv.org)

Does an Agent's History Tell You When Compaction Will Hurt? A Modest, Bounded Effect on the TRACE Paired-Replay Corpus (arxiv.org)

Toward Alignment Scaling Laws: A Framework and First Preregistered Measurements (arxiv.org)

Beyond Perturbation Magnitude: Direction-Dependent Responses in Multimodal Geometric Representations (arxiv.org)

Incentive Alignment in Online Experimentation (arxiv.org)

X-OPM: Explainable Automatic Digital On-Chip Power Modeling for Enhanced Robustness (arxiv.org)

UNREAL: Unifying Retrieval and Long-Context with a Single Model (arxiv.org)

WorldSolver: Can LLM Agents Simulate the Physical Dynamics via Solver Generation? (arxiv.org)

nanoMuse: An Open-Source Personal Agent for Every Device You Own (arxiv.org)

ScienceClaw: Benchmarking Continual Self-Evolution of AI-for-Science Agents Across the Natural and Social Sciences (arxiv.org)

Coupled but Late: Turn-Taking Between Full-Duplex Speech Models in Unscripted Dialogue (arxiv.org)

ParanoiaEval: Benchmarking Unnecessary Defensive Work in Agentic Coding (arxiv.org)

SquidAgent: Parallelize Wisely, Coordinate Efficiently (arxiv.org)

Parallel Predictive World Models for Accurate and Efficient Long-Horizon Planning (arxiv.org)

Personal-Agent Mediated Recommendation with Cross-Platform User History (arxiv.org)

Representation Bias, Correction Transfer, and Resolution Sensitivity in Three-Dimensional Mitochondrial Morphometry (arxiv.org)

Agentic AutoRAG: RAG Pipeline Optimization through Reasoning-Driven Agents (arxiv.org)

Knowing When Not to Answer: Cross-Domain and Multi-Turn Generalization of Latent Underspecification Signals (arxiv.org)

DIPrune: Task-Aware Token Pruning with Dual Importance for Efficient Multimodal Language Models (arxiv.org)

Machine Learning for German Redispatch Forecasting under Data Delays and Temporal Distribution Shift (arxiv.org)

The Standardization Trap: Certifying Joint Label Processing in Tabular Foundation Models (arxiv.org)

Cooperating with Future Collaborators: Multi-Agent RL under Staggered Participation (arxiv.org)

Unanimously Wrong: Certified Abstention from How Medical LLM Consensus Forms (arxiv.org)

Navigating Route Latent Space for Synthesizable Molecular Design (arxiv.org)

A Systematic Investigation of Bias in Large Language Models for Advertising Relevance (arxiv.org)

Grounding What Shapes the Plan: Rethinking Groundedness for Physical Intelligence in Autonomous Driving (arxiv.org)

CoDe-LoRA: Mitigating the Orthogonality Dilemma in Continual Learning of LLMs via Knowledge Consolidation and Decoupling (arxiv.org)

Where Do Two Populations of Persistence Diagrams Differ? Calibrated Local Inference at a Fixed Budget (arxiv.org)

DySCo: Dynamic Sharding for Collaborative Edge-Cloud LLM Inference with Depth-Synchronized Batching (arxiv.org)