AI/ML arXiv cs.AI

AI-PAVE-Br: Leveraging Large Language Models for Enhanced Product Attribute Value Extraction through a Golden Set Approach

AI-PAVE-Br is introduced as a specialized LLM-based system for product attribute value extraction in Brazilian e-commerce, accompanied by a new Portuguese benchmark dataset.

AI/ML arXiv cs.AI

FlowPipe: LLM-Enhanced Conditional Generative Flow Networks for Data Preparation Pipeline Construction

FlowPipe is a framework that uses Conditional Generative Flow Networks and LLM-derived priors to automatically construct optimized data preparation pipelines.

AI/ML arXiv cs.AI

TACTFUL: Tactile-Driven Exploration For Object Localization and Identification in Confined Environments

TACTFUL is a vision-free tactile exploration framework enabling robots to autonomously locate and identify objects in confined spaces using tactile reconstruction.

AI/ML arXiv cs.AI

Evaluating the Interpretability of Sparse Autoencoders with Concept Annotations

A human-grounded evaluation framework is presented to quantify the alignment between Sparse Autoencoder latents and human-annotated concepts in vision models.

Other Hacker News

Blogging Can Just Be Stating the Obvious

A discussion on the nature of blogging and the value of stating the obvious in written communication.

Tech Business/VC Hacker News

Anthropic says Alibaba illicitly extracted Claude AI model capabilities

Anthropic accuses Alibaba of illicitly extracting capabilities from the Claude AI model.

Software Engineering Hacker News

LuaJIT 3.0 proposed syntax extensions

Discussion regarding proposed syntax extensions for LuaJIT 3.0.

Other Hacker News

Dostoyevsky isn't difficult

A discussion about the accessibility and difficulty of reading Dostoyevsky's works.

AI/ML arXiv cs.AI

G$^3$VLA: Geometric inductive bias for Vision-Language-Action Models

Introduction of G3VLA, a camera-aware geometric module for Vision-Language-Action models to improve robot manipulation using calibrated geometry.

AI/ML arXiv cs.AI

video-SALMONN-R$^3$: Learning to ReWatch, ReAsk, and ReAnswer for Efficient Video Understanding

video-SALMONN-R3 presents an end-to-end video-LLM using reinforcement learning for efficient video understanding through a re-watch and re-answer strategy.

AI/ML arXiv cs.AI

Adaptive Machine Learning Framework for UAV Trajectory Optimization in O-RAN

A new machine learning framework for optimizing UAV trajectories in O-RAN 6G systems using continual transfer learning.

AI/ML arXiv cs.AI

RetiSEM: Generalising Causal Models for Fragmented Biomedical Data

RetiSEM is a structural equation modelling framework for recovering causal graphs from fragmented biomedical data.

Cybersecurity arXiv cs.AI

Red-Teaming the Agentic Red-Team

An analysis of security flaws in agentic offensive security tools, proposing a robust architecture to mitigate common attack paths.

AI/ML arXiv cs.AI

CrossPool: Efficient Multi-LLM Serving for Cold MoE Models through KV-Cache and Weight Disaggregation

CrossPool is a serving engine for cold MoE models that disaggregates FFN weights and KV-cache into separate GPU memory pools for efficiency.

Hardware/Chips TechCrunch

Europe is pushing back on Washington’s chip war

Europe is resisting U.S. chip export restrictions to China, specifically regarding older-generation deep ultraviolet lithography tools.

AI/ML arXiv cs.AI

Detecting AI Coding Agents in Open Source: A Validated Multi-Method Census of 180 Million Repositories

A large-scale study of 180 million repositories reveals that AI coding agents like Claude Code are significantly more prevalent than previous bot-signature methods suggested.

AI/ML arXiv cs.AI

Transformation Behavior of Images in Latent Space

Research examines how classical image transformations affect the latent space of encoder networks in histopathology classification, finding they are robust but not fully invariant.

AI/ML arXiv cs.AI

MedPCFM: Improving Medical Point Cloud Completion by Integrating Point Transformers and Flow Matching

MedPCFM introduces a flow matching approach using Point Transformers (PTv3) to improve medical point cloud completion for anatomical reconstruction.

AI/ML arXiv cs.AI

NoContactNoWorries: Estimating Contact through Vision and Proprioception for In-Hand Dexterous Manipulation

NoContactNoWorries is a multimodal framework that allows robots to infer physical contact using vision and proprioception instead of dedicated tactile sensors.

AI/ML arXiv cs.AI

The African Language Tax: Quantifying the Cost, Latency, and Context Penalty of Tokenizing African Languages in Frontier LLMs

A study quantifies the 'African Language Tax,' showing that frontier LLMs charge significantly more in cost and latency for African languages due to inefficient tokenization.