Signal over volume.

Algorithm Audit Reveals LLMs Overweight Eco-Certification and Ignore Manageme...

Algorithm Audit Reveals LLMs Overweight Eco-Certification and Ignore Management Response in Hotel Selection

Frontier AI Systems Out-Persuade Expert Humans in Attitudinal and Behavioral ...

Frontier AI Systems Out-Persuade Expert Humans in Attitudinal and Behavioral Contests

EARS: Early Scam Scenario Detection via Conversation-Aware Modeling

REDACT: Protecting Proprietary Agent Skills via Selective Trace Rewriting and...

Avatar V: Scaling Video-Reference Identity and Motion Modeling for Production...

OmniDirector: Achieving Precise Multi-Shot Camera Motion Cloning via Camera G...

RhymeFlow: Accelerating Video Generation via Asynchronous Denoising Flow Sche...

LoRA-α: Unlocking Low-Rank Adaptation via Principled Scaling Factor Optimization

LLM Agents Can See Code Repositories: Multimodal Context Boosts Efficiency

HarnessX: A Foundry for Composable and Evolving AI Agent Runtime Interfaces

Hy-Embodied-0.5-VLA: A Full-Stack Robot Learning System for Real-World Deploy...

MBench: A New Hierarchical Benchmark for Evaluating Memory in Video World Models

POLAR: Dynamically Skipping or Looping LLM Layers for Efficient Reasoning

SteerBoost: Predicting LLM Activation Steering Success from Early Hidden States

Orchestra-o1: A Scalable Framework for Omnimodal Agent Orchestration

µ0: A Scalable 3D Interaction-Trace World Model for Cross-Embodiment Robot Le...

VISTA: Enhancing GUI Grounding via View-Consistent Self-Verified Training

LLM Annotation Reliability Limited by Internalized Priors and Decision Sticki...

Generative AI Exposure as a Marker of Caste Privilege in India's Graduate Lab...

StakeBench: A Stakeholder-Centric Benchmark for Prompt Injection in Web Agents

AiAWE: Open-Source LLM System Outperforms Proprietary Models in Essay Scoring

Neural-Network Chess Engines Drive Global Increase in Draw Rates Without Chan...

Bebop: Breaking Entropy Bounds to Accelerate RL Training via MTP and TV Loss

ICALens: A Practical Workflow for Interpreting LLMs via Independent Component...