All Articles
16560 articles total
Competitive and Complementary Tools
A theoretical model analyzing the bistable relationship between human competence and tool reliance, warning of competence collapse when tools are too available.
RRPO: Reference-Relative Policy Optimization with Stratified Conditional Rollouts
Introduces Reference-Relative Policy Optimization (RRPO), a generalization of GRPO that uses contrastive comparisons instead of ground-truth verifiers for RL alignment.
Never Enough
A Hacker News discussion thread titled "Never Enough", primarily consisting of user comments.
Addressing Limited Data in Auditory Attention Decoding with Diffusion Generative Models
Researchers use diffusion probabilistic models (DPMs) to generate synthetic EEG data to improve Auditory Attention Decoding (AAD) for hearing aids.
An Analysis of Residual-Stream Geometry Across Transformer Depth
A geometric analysis of transformer residual streams revealing reproducible depth regularities in instruction-tuned models.
MambaLSTM: A Spatio-Temporal Framework for Enhanced Traffic Accident Risk Prediction
Introduction of MambaLSTM, a spatio-temporal framework using state-space models to enhance traffic accident risk prediction.
Multi-layer MIMO Relay as Deep Physical Neural Networks: Power Amplifiers as Activation Functions
A proposal for wireless physical neural networks (WPNNs) that use MIMO relay networks and power amplifiers as activation functions for over-the-air inference.
CODENS: Transforming Code Changes into Living, Accessible, and Queryable Documentation
CODENS is a system that converts pull requests into a queryable knowledge graph to create living documentation for production codebases.
Decode-Time Grammars: Constrained LLM Generation over a Refinement Order of Grammar Fragments
Introduction of decode-time grammars to ensure both grammatical and semantic correctness in LLM-generated code for low-resource languages.
HALLMARK: Diagnosing Three Failure Modes in LLM Citation Verifiers
Introduction of HALLMARK, a benchmark for diagnosing failure modes in LLM citation verifiers, highlighting the criticality of false-positive rates.
Physical Self-Supervised Learning: IMU Sensing without Manual Labels
A new physical self-supervised learning paradigm for label-free IMU sensing using an auto-adaptive physics decoder.
A Controlled Study of Attention-Only Transformers
A study on attention-only transformers (SANs) showing they can match standard transformers in many aspects but struggle with knowledge-dense tasks.
Meta made its own AI detection system. It should have just used Google’s
Meta introduced Content Seal, an invisible watermarking system for AI-generated images, though critics argue it is less reliable than existing industry standards like C2PA.
The 2026 Honda Prelude is a marvel of hybrid technology
A look at the technological evolution of the Honda Prelude, highlighting its history of integrating hybrid and engineering innovations.
Hazard or Anomaly? Evaluating VLMs for Understanding Dangers and Discrepancies
Researchers propose a new evaluation framework for Vision-Language Models (VLMs) that distinguishes between true physical hazards and mere anomalies to improve safety reasoning.
Dynamic Loss Balancing for Joint SOH and RUL Prediction of Lithium-Ion Batteries via a Rotary SOH-Injected Prior Battery Transformer
The RoSIP-Batt framework uses a Rotary SOH-Injected Prior Battery Transformer to jointly predict the State of Health and Remaining Useful Life of lithium-ion batteries.
Physics-Guided Masked Multi-Task Network for Edge-Friendly Battery Health Diagnostics from Sto-chastically Fragmented Charging Profiles
The RoSIP-Batt framework uses a Rotary SOH-Injected Prior Battery Transformer to jointly predict the State of Health and Remaining Useful Life of lithium-ion batteries.
ChemHyperMag: Physics-informed magnetic hypergraph learning improves molecular ADMET prediction
ChemHyperMag introduces physics-informed magnetic hypergraph learning to improve the prediction of ADMET properties in drug discovery.
Quantum Cryptanalysis on IBM Quantum Hardware: Extending Even--Mansour Period Recovery from $N=4$ to $N=10$
Researchers demonstrate real-hardware quantum cryptanalysis of the Even-Mansour cipher on IBM Quantum hardware, extending key recovery to larger security parameters.
PRISM: Sensitivity-Aware PolynoMial PRuning for EffIcient Neural Network Encryption
PSAP is introduced as a reliability-aware structured pruning method for neural networks under homomorphic encryption, reducing vulnerability to bit-flip injections.