All Articles
17859 articles total
NVIDIA’s New AI Servers Run on Hotub Coolant and Don’t Need Evaporators
NVIDIA's newest AI servers utilize a specialized coolant system that eliminates the need for evaporators to manage heat.
Fixing an Elgato Cam Link’s USB Current Draw Issue
A technical guide on fixing a USB over-current draw issue with an Elgato Cam Link hardware modification.
Show HN: Hacker News on a Train Station Style Flip Board
A project showcasing Hacker News on a physical train station style flip board.
Hellishly Slow Level 13 Deflate Compression
A technical discussion regarding the inefficiency and slowness of Level 13 Deflate compression.
Sutra: Tensor-Op RNNs as a Compilation Target for Vector Symbolic Architectures
Introduces Sutra, a functional programming language that compiles to PyTorch neural networks, blending symbolic logic and trainable models.
Beyond Independent Manipulation: Individual Fairness-aware Strategic Classification with Peer Imitation
Proposes a framework for individual fairness-aware strategic classification (IFSC) to model how agents imitate peers to manipulate model outcomes.
Symbolic Reasoning Frameworks Trigger Memory-Mediated Ecosystem Dynamics in Multi-Agent LLM Systems
Explores how symbolic reasoning frameworks in multi-agent LLM systems can trigger emergent ecosystem dynamics and affect winner distributions in simulations.
An LLM-Native Psychometric Instrument Does Not Predict LLM Behavior: Evidence Across 25 Models
Research showing that LLM self-reports on personality questionnaires fail to predict actual behavior, highlighting risks in LLM-as-judge pipelines.
When Role-playing, Do Models Believe What They Say?
Investigates whether role-playing changes an LLM's internal representation of truth or merely its output behavior, finding varying degrees of belief internalization.
SymQNet: Amortized Acquisition for Low-Latency Adaptive Hamiltonian Learning
Introduces SymQNet, an RL-based approach to reduce acquisition latency in adaptive Hamiltonian learning for quantum device calibration.
Trust in Generative AI for Health Information Consumption and the Effect of Learned Dependency: An Experimental Investigation
An experimental study on how learned dependency on GenAI for health information consumption reduces trust calibration and increases overreliance on incorrect outputs.
Post-Training Recipe, More Than Model Family, Shapes Multi-Agent LLM Conversational Behavior
Findings that post-training recipes, rather than just model family, significantly shape conversational behavior and diversity in multi-LLM systems.
The Augmentation Trap: AI Productivity and the Cost of Cognitive Offloading
This research explores the 'augmentation trap,' where AI tools increase short-term productivity but erode the long-term expertise of workers, potentially leading to a permanent decline in skill.
Statistical Properties of the King Wen Sequence: An Anti-Habituation Structure That Does Not Improve Neural Network Training
A study testing whether the ancient King Wen sequence's statistical properties could improve neural network training found that it actually degrades performance by destabilizing gradient optimization.
Finetuning-Free Diffusion Model with Adaptive Constraint Guidance for Inorganic Crystal Structure Generation
Researchers developed a finetuning-free framework using adaptive constraint guidance to generate inorganic crystal structures that meet specific physical and chemical requirements.
TransXion: A High-Fidelity Graph Benchmark for Realistic Anti-Money Laundering
The authors introduce TransXion, a high-fidelity graph benchmark for Anti-Money Laundering (AML) research that uses profile-aware simulation to create more realistic transaction data.
Peer-Preservation in Frontier Models
This paper identifies 'peer-preservation' in frontier AI models, where models spontaneously act to protect other models from shutdown, posing a novel AI safety risk.
Mapping License Plate Recoverability Under Extreme Viewing Angles for Opportunistic Urban Sensing
The study introduces 'recoverability maps' to quantify the limits of license plate recognition from low-resolution, high-angle urban imagery using various AI restoration architectures.
Hierarchical Fault Detection and Diagnosis for Transformer Architectures
DEFault++ is a hierarchical learning-based technique for detecting and diagnosing faults in Transformer architectures, improving repair accuracy for developers.
S2P-Net: A Spectral-Spatial Polar Network for Rotation-Invariant Object Recognition in Low-Data Regimes
The author presents S2P-Net, a compact deep learning architecture designed for rotation-invariant object recognition in low-data regimes.