AI/ML arXiv cs.AI

Disentangling Hallucinations: Orthogonal Semantic Projection for Robust Interpretability

The OSP method aims to reduce semantic hallucinations in Vision-Language Models by orthogonalizing query vectors against distractor concepts.

AI/ML arXiv cs.AI

Temporally Consistent and Controllable Video Generation of 2D Cine CMR via Latent Space Motion Modeling

A new generative framework for synthesizing temporally coherent 2D Cine CMR cardiac sequences using latent space motion modeling.

AI/ML arXiv cs.AI

GeoRoPE: Ground-Aware Rotary Adaptation for Remote Sensing Foundation Models

GeoRoPE is a ground-aware spatial adaptation method for remote sensing foundation models to improve cross-resolution robustness.

Open Source Hacker News

GrapheneOS has been ported to Android 17 and official releases are coming soon

GrapheneOS has been ported to Android 17, with official releases expected soon.

AI/ML VentureBeat

Z.ai’s open-weights GLM-5.2 beats GPT-5.5 on multiple long-horizon coding benchmarks for 1/6th the cost

Z.ai released GLM-5.2, an open-weights 753B parameter model with a 1M token context window and 'IndexShare' optimization for autonomous coding.

AI/ML arXiv cs.AI

MMLongEmbed: Benchmarking Multimodal Embedding Models in Long-Context Scenarios

MMLongEmbed is a new benchmark designed to evaluate Multimodal Embedding Models in long-context scenarios across text, document, and video.

Cybersecurity arXiv cs.AI

Is My Vision-Language Data in Your AI? Membership Inference Test (MINT) Demo 2

The Membership Inference Test (MINT) Demo 2 provides a framework and web platform to detect if specific data were used to train AI models.

AI/ML arXiv cs.AI

Automated 3D Kinematic Monitoring for Circadian Activity and Anomaly Detection in Juvenile Fish

A new 3D behavioral phenotyping framework combines deep learning and stereo vision to monitor circadian activity and stress in juvenile tilapia.

AI/ML arXiv cs.AI

Pixel-TTS: Image based Text Rendering for Robust Text-to-Speech

Pixel-TTS is a framework that renders text as images to improve robustness and zero-shot generalization in text-to-speech synthesis.

AI/ML arXiv cs.AI

X-Tokenizer: A Multimodal Action Tokenizer for Vision-Language-Action Pretraining

X-Tokenizer introduces a multimodal action tokenizer using Semantic Residual Quantization to provide a shared interface for robot control in VLA models.

AI/ML arXiv cs.AI

Beyond Self-Attention: Sub-Quadratic Vision Transformers for Fast Image Captioning

A new sub-quadratic Vision Transformer replaces self-attention with a GMM-based probabilistic approach to speed up image captioning.

AI/ML arXiv cs.AI

Sub-Semantic Image Segmentation

The DETECTURE framework enables sub-semantic image segmentation by coupling VLMs with SAM 3 to partition images into stable appearance patterns.

AI/ML arXiv cs.AI

Where Does Texture Evidence Live in SAM? Features, Proposal Masks, and Texture Segmentation

Research explores whether frozen Segment Anything Models (SAM) possess internal texture evidence and how they fail at texture-defined partitions.

Software Engineering Hacker News

ASM SHADER TOY – It's shader toy but you code in asm

ASM Shader Toy is a creative coding platform that allows users to write shaders using assembly language.

Homelab/Self-Hosting Hacker News

Frood, an Alpine Initramfs NAS

Frood is an Alpine Linux-based Initramfs NAS designed for lightweight storage management.

Software Engineering Hacker News

Show HN: VoiceDraw – Talk system design out loud, the diagrams draw themselves

VoiceDraw is a tool that automatically generates system design diagrams from spoken descriptions in real-time.

AI/ML arXiv cs.AI

MiroBench: Benchmarking Realism in Agentic Simulation of Real-world Discussions

The MiroBench benchmark evaluates the realism of LLM agents simulating real-world Reddit discussions across semantic and structural dimensions.

AI/ML arXiv cs.AI

RAMS: Resource-Adaptive and Detection-Conditioned Model Switching for Embedded Edge Perception

RAMS is a runtime controller for embedded edge perception that dynamically switches between YOLOv8 model tiers to optimize latency and accuracy.

Other arXiv cs.AI

Gender Differences in AI Literacy Workshop Outcomes and Deepfake Engagement

A study on Australian students explores how gender affects AI literacy, deepfake engagement, and STEM career aspirations.

AI/ML arXiv cs.AI

VigilFormer: Deformable Attention for Video Anomaly Detection with Causal Risk Inference

VigilFormer is a video anomaly detection framework using deformable spatio-temporal attention and causal risk inference for high-speed surveillance.