Signal over volume.

Large-scale study reveals systematic L1 bias in fine-tuned LLM automated essa...

Large-scale study reveals systematic L1 bias in fine-tuned LLM automated essay scoring

Do Generative AI Assistants Respect robots.txt? An Empirical Study of Web Acc...

Do Generative AI Assistants Respect robots.txt? An Empirical Study of Web Access Compliance

Finetuning on Benign Data Triggers Broad Ideological Shifts in LLMs

Multi-Agent Debate Fails to Outperform Single-Pass AI for Research Paper Feed...

AI advice suppresses human willingness to say 'I don't know', even when advic...

AI-Native Insurance Framework for Agentic AI: Pricing, Underwriting, and Auto...

Generalized Nonlinear Imaginary-Time Evolution (NITE) for Advanced Quantum St...

New Si/SiGe Heterostructure Design Pushes Valley Splitting to 5 meV for Scala...

Study Reveals Significant Voting Biases in DAO Governance via Interface and A...

L2-Bench: A New Pedagogical Benchmark for Evaluating LLMs in Second Language ...

Privacy Detective: A Narrative Game Training Developers to Spot Real-World Pr...

AI Guardrails Shift Healthcare Decisions: Field Experiment Reveals Directiona...

Diagnosing and Repairing Persona Collapse in LLM Advice

Pluralis v0.1: A New Multimodal, Multilingual Benchmark for Localized AI Safe...

Algorithmic Pricing: How Information Disclosure Rules Shape Collusion and Pro...

Strategic Buying Agents: Optimizing Purchase Timing Under Price Uncertainty

Schwartz-Geometry Decoding: Making Human Value Detection Faithful to Psycholo...

Dispute Content Shapes Public Acceptance of AI in Legal Adjudication

Transition-metal NiV− defects in diamond enable all-optical control and milli...

SPLIT Benchmark Reveals LLMs Struggle with Ukrainian Cultural Empathy

Synthetic Contact with AI Chatbots Reduces Cross-Partisan Animosity and Mispe...

Eticas AI Risk Taxonomy v2.0.0: Bridging the Gap Between Risk Cataloging and ...

DART: One-Shot VLA Adaptation via Domain Arithmetic and Subspace Alignment

DiscoPER: An Autonomous Framework for Open-Ended Multimodal Scientific Discovery