AI/ML arXiv cs.AI

Assessing VLM Reliability for Medical Image Quality Evaluation Under Corruption and Bias

A study on the reliability of Vision-Language Models (VLMs) for medical image quality assessment, highlighting sensitivities to corruption and bias.

Cybersecurity Hacker News

MSI Center – How to gain SYSTEM privileges in seconds

A security vulnerability in MSI Center allows users to gain SYSTEM privileges on Windows in seconds.

AI/ML Hacker News

CueBench for Developers is live: score how well you drive coding agents

CueBench for Developers is launched to help developers evaluate the effectiveness of their coding agents.

Other Hacker News

Scientists discover guidance system for migratory songbirds

Scientists have identified the guidance system used by migratory songbirds.

Software Engineering Hacker News

Software, from First Principles

An exploration of software development based on first principles.

Software Engineering Hacker News

Notes from Building Tinkerfont

A detailed account of the process of building Tinkerfont.

Software Engineering Hacker News

You can get Unicode working on DOS

A technical explanation of how to achieve Unicode support on DOS.

AI/ML arXiv cs.AI

SABER: A Semantic-Aligned Brain Network Analysis Framework via Multi-scale Hypergraphs

SABER is a new framework that uses multi-scale hypergraphs and LLM-derived semantics to improve brain disease diagnosis.

AI/ML arXiv cs.AI

Population-Based Multi-Objective Training of Discriminators for Semi-Supervised GANs

A new population-based evolutionary training strategy for semi-supervised GANs to improve training stability and classification accuracy.

AI/ML arXiv cs.AI

Low-Latency Task-Oriented Image Transmission with Opportunistic Spectrum Access

A transmission framework using VQ-VAE and opportunistic spectrum access to achieve low-latency task-oriented image transmission.

AI/ML arXiv cs.AI

TUDUM: A Turkish-Thinking Reasoning Pipeline for Qwen3.5-27B

TUDUM is a pipeline for adapting Qwen-family models to perform reasoning and internal thinking specifically in Turkish.

AI/ML arXiv cs.AI

Expander Sparse Autoencoders: Parameter-Efficient Dictionaries for Mechanistic Interpretability

Researchers introduce Expander Sparse Autoencoders (SAEs) to significantly reduce the parameter count and storage requirements of dictionaries used in mechanistic interpretability.

Software Engineering arXiv cs.AI

Decoupling Code Complexity from Newcomer Participation: A Causal Study of AI Coding Agent Adoption in OSS

A causal study finds that the adoption of AI coding agents in open-source projects increases code complexity but does not significantly crowd out newcomer participation.

AI/ML arXiv cs.AI

MMBench-Live: A Continuously Evolving Benchmark for Multimodal Models

MMBench-Live is presented as a continuously evolving multimodal benchmark for vision-language models to prevent data contamination and temporal staleness.

AI/ML arXiv cs.AI

Mixture-of-Parallelisms: Towards Memory-Efficient Training Stack for Mixture-of-Experts Models

Mixture-of-Parallelisms (MoP) is a memory-efficient training stack for MoE models that enables trillion-parameter scale pre-training with significantly higher throughput than FSDP2.

AI/ML arXiv cs.AI

Decomposer: Learning to Decompile Symbolic Music to Programs

Decomposer is a framework that uses reinforcement learning to decompile symbolic MIDI music into readable and executable programs in the Strudel language.

AI/ML arXiv cs.AI

Evaluating Chunking Strategies for Retrieval-Augmented Generation on Academic Texts

An evaluation of RAG chunking strategies on academic texts suggests that complex cluster-based semantic chunking does not necessarily outperform simpler fixed-size or recursive strategies.

Cybersecurity arXiv cs.AI

Has This Checkpoint Been Abliterated? A Two-Signal Audit and Its Failure Map

A new audit method combines activation gaps and weight-recovery energy to detect if an open-weight LLM checkpoint has had its refusal mechanisms removed (abliterated).

Software Engineering arXiv cs.AI

An Exploratory Study on LLM-Generated Code and Comments in Code Repositories

A study of LLM-generated code in repositories suggests that such code is frequently found in test cases and exhibits high levels of internal cloning, with little direct association with bugs.

AI/ML arXiv cs.AI

SAB-LVLM: Significance-Aware Binarization for Large Vision-Language Models

SAB-LVLM introduces significance-aware binarization for Large Vision-Language Models to reduce memory and latency for resource-constrained deployment.