AI/ML arXiv cs.AI

Whose fairness? Structural concentration in AI bias research

Analysis of AI bias research indicates structural concentration, with a small set of Western institutions dominating the definitions and benchmarks used globally.

Other TechCrunch

Venus Williams-backed WeWard can now lock your apps until you hit your steps

WeWard is a French app that incentivizes walking by locking other apps until users reach their daily step goals.

Hardware/Chips The Verge

First look: Fi Ultra Starlink pet tracker

Fi Ultra is a new pet tracker that utilizes Starlink's direct-to-cell satellite service for connectivity in cellular dead zones.

AI/ML arXiv cs.AI

Decision Protocols in Multi-Agent Large Language Model Conversations

Research introduces the Multi-Agent LLM (MALLM) framework to evaluate decision protocols like voting and consensus for improving task performance in multi-agent AI systems.

AI/ML arXiv cs.AI

Privilege and confidentiality in generative AI workflows

An analysis of data storage and processing modes in generative AI systems and their implications for legal professional privilege and confidentiality.

AI/ML arXiv cs.AI

Full-range Binary Classifier Calibration for Stable Model Updates in Production

A new method for full-range binary classifier calibration is proposed to maintain consistent false-positive rates during stable model updates in production environments.

AI/ML arXiv cs.AI

PatchOptic for Shared-State LLM Workflows with Projected Views and Verified Structured Updates

PatchOptic is introduced as an optic-inspired interface to manage shared-state LLM workflows using projected reads and verified structured patches.

AI/ML arXiv cs.AI

Lean-Quantum: Toward AI-Assisted Formalization of Quantum Information

The Lean-Quantum library provides a formal infrastructure in Lean 4 for AI-assisted formalization of quantum information theory.

AI/ML arXiv cs.AI

Statistical Adversaries: Natural Backdoor-like Features in Vision Datasets

Research identifies 'statistical adversaries' in vision datasets—natural patterns that behave like backdoor triggers without being maliciously inserted.

Cybersecurity arXiv cs.AI

aiAuthZ: Off-Host, Identity-Bound Authorization for AI Agents

aiAuthZ is an authorization gateway that prevents AI agents from acting beyond verified user authority using identity-bound tool calls.

AI/ML arXiv cs.AI

Rendering-Aware Bayesian 3D Gaussian Splatting with Native Uncertainty and Adaptive Complexity Control

A rendering-aware Bayesian 3D Gaussian Splatting framework is introduced to provide native uncertainty and adaptive complexity control for novel-view synthesis.

Hardware/Chips Hacker News

Apple to increase spend with Broadcom to produce billions more U.S. chips

Apple is increasing its spending with Broadcom to significantly ramp up the production of chips manufactured in the United States.

Other Hacker News

How to Survive 3 Years in North Korea as a Foreigner

A personal account detailing the experience of living as a foreigner in North Korea for three years.

Tech Business/VC TechCrunch

Former OpenAI exec Kevin Weil is now on the board of Stoke Space

Former OpenAI executive Kevin Weil has joined the board of Stoke Space, signaling a trend of AI leadership moving into the reusable rocket industry.

Hardware/Chips The Verge

The Steam Machine fits my TV, my desk, and my life

A review of the Steam Machine, highlighting its ability to integrate a Steam library across different devices despite some drawbacks.

AI/ML VentureBeat

Slack’s Slackbot can now pull your CRM data, generate charts, and send DocuSigns — all from a chat message.

Slack integrates Salesforce CRM data and Tableau analytics into Slackbot using the Model Context Protocol (MCP), enabling 'multiplayer AI' orchestration.

AI/ML arXiv cs.AI

Learning 4D Geometric Priors for Inference-Efficient World Action Models

MECo-WAM is introduced to improve robotic manipulation by injecting 4D geometric priors into video-action representations without increasing inference cost.

AI/ML arXiv cs.AI

Breaking Structural Isolation: Scalable Graph Clustering via Community-Aware Sampling and Structural Entropy

The SCISE framework is proposed for scalable unsupervised graph clustering, using community-aware sampling and structural entropy to maintain topological integrity.

AI/ML arXiv cs.AI

KAT-Coder-V2.5 Technical Report

KAT-Coder-V2.5 is an agentic coding model trained in executable repositories using an end-to-end post-training framework for autonomous software engineering.

AI/ML arXiv cs.AI

Binocular Gaze Estimation with Single Camera and Single Light Source

A new gaze estimation method is proposed using only one camera and one light source by introducing a 'virtual light source' and virtual glints.