Homelab/Self-Hosting Ars Technica

I wanted a clock that never needed setting. Things escalated.

A detailed account of a hobbyist project to build a clock that never needs manual setting, involving a deployment pipeline.

AI/ML arXiv cs.AI

Reinforcement Learning for Large Language Model Selective Evidence Adoption from Contaminated Retrieval Results

Introduces SelectBench, a benchmark for evaluating LLMs' ability to selectively adopt evidence from contaminated retrieval results, and tests it using DAPO.

AI/ML arXiv cs.AI

ENTRAP-VL: A Taxonomic Probe for Dual Contextual Entrainment in Vision-Language Models

Presents ENTRAP-VL, a taxonomic probe designed to investigate contextual entrainment in vision-language models across textual and visual streams.

AI/ML arXiv cs.AI

PRIME-SVR: Physics-infoRmed Implicit Multi-Echo Slice-to-Volume Reconstruction for Fetal T2 mapping

Introduces PRIME-SVR, an implicit neural representation framework for high-resolution 3D fetal brain reconstruction from multi-echo MRI.

Hardware/Chips arXiv cs.AI

Formal Foundations for Known Good Reliable Die Screening in Chiplet-Based AI Systems-on-Chip

Proposes a formal framework for 'Known Good Reliable Die' (KGRD) screening to improve the reliability of chiplet-based AI systems-on-chip.

AI/ML arXiv cs.AI

SLAI T-Rex: Full-Parameter Post-training of the DeepSeek-V4 Family on Ascend SuperPOD

Details SLAI T-Rex, a system for efficient full-parameter post-training of trillion-parameter MoE models on Ascend NPU SuperPODs.

Hardware/Chips Hacker News

New Framework Desktop Option with AMD Ryzen AI Max+ Pro 495 and 192GB Memory

Framework has announced a new desktop configuration featuring the AMD Ryzen AI Max+ Pro 495 processor and 192GB of memory.

AI/ML arXiv cs.AI

TINY_SCHILLER: A Drop-In German Drama Corpus for Small Language Models

TINY_SCHILLER is a 2.07MB German drama corpus designed as a drop-in replacement for tiny_shakespeare for small language model research and education.

AI/ML arXiv cs.AI

Are Attributions of Consciousness to AI Chatbots Epistemically Innocent?

A conceptual analysis and taxonomy exploring whether attributing consciousness to AI chatbots is epistemically justified or an irrational belief.

AI/ML arXiv cs.AI

Post-Training in Time Series Foundation Models: A Unifying Framework

A proposed unifying framework for post-training Time Series Foundation Models (TSFMs) to bridge the gap between pretraining and downstream deployment.

Cybersecurity arXiv cs.AI

Taming the Security-Energy Paradox: A Green AI Approach to Optimized Android Malware Detection

Research demonstrating that INT8 quantized neural networks can maintain 99.2% accuracy in Android malware detection while significantly reducing energy consumption.

Cybersecurity arXiv cs.AI

Drift-Aware RL-based Wavelet Denoising for Network-Traffic Anomaly Detection

A drift-aware framework using RL-based wavelet denoising to improve network-traffic anomaly detection and capacity estimation.

AI/ML arXiv cs.AI

A Systematic Benchmark of Intensity Normalisation Methods for 3D Knee MRI Segmentation and Cross-Domain Generalisability

A study comparing intensity normalization methods for 3D knee MRI segmentation, finding that domain shift has a larger impact on generalizability than normalization.

AI/ML arXiv cs.AI

Test Case Prioritization for DNNs via Neural Collapse Instability

The NCIP framework improves test case prioritization for DNNs by using prediction variability across checkpoints rather than absolute confidence scores.

AI/ML arXiv cs.AI

Language-Specific versus Cross-Lingual Knowledge Graphs for Implicit Aspect Identification in Arabic: A Comparative Study of Reasoning and Adaptation Strategies

Comparative study showing that native Arabic knowledge graphs and task-specific fine-tuning outperform cross-lingual English KGs for implicit aspect identification in Arabic.

AI/ML arXiv cs.AI

Co-Evolving LLM Evaluators and Policies via DynamicRubric

DynamicRubric is a co-evolution framework for LLM evaluators and policies that improves reasoning and coding tasks by evolving the evaluation criteria alongside the model.

Software Engineering Hacker News

Code mode yields a 99.2% cost reduction in our systems

An article discussing a 'Code mode' that achieved a 99.2% cost reduction in system operations.

Software Engineering Hacker News

The Unity CLI: manage Unity from your terminal

Introduction of a Command Line Interface (CLI) for Unity to allow management of the game engine from the terminal.

Other Hacker News

Worse on Purpose – How Corporate Greed Killed Product Quality – Worse on Purpose

A critique of corporate greed and its negative impact on product quality, framed as 'Worse on Purpose'.

AI/ML TechCrunch

Experts say exploiting Anthropic’s Fable isn’t how Kimi K3 got so good

Experts debate whether Kimi K3's performance gains are primarily due to distillation from Anthropic's Fable model.