All published AI news and analysis.

Omadia is a self-hostable agentic operating system designed for building, running, and auditing multi-agent AI teams. It supports signed plugins, allows users to bring their own LLM keys, and emphasizes data ownership with EU/GDPR readiness.

DeepSeek has announced the release of DeepSeek-V4-Flash-Vision-Exp. This update pertains to their vision model capabilities.

A new study compares ceiling-mounted FMCW, IR-UWB, and Wi-Fi radar for in-bedroom human activity recognition (HAR) and sleep monitoring. IR-UWB led in cross-subject HAR (89.0% macro F1), while FMCW excelled in generalizing to new room layouts (83.8% macro F1). All technologies achieved over 92% macro F1 for sleep monitoring.

Researchers have introduced ConceptGuard, a new benchmark designed to evaluate the context-sensitive unlearning capabilities of Large Language Models (LLMs). Existing benchmarks often fail to capture the nuanced requirement of removing harmful knowledge while preserving beneficial information.

Google DeepMind is collaborating with game studios to develop and test advanced AI gameplay prototypes. This initiative builds upon their extensive 15-year history of AI research in the gaming sector.

Researchers have introduced DreamHand, an offline framework that repurposes video diffusion models (VDMs) as deterministic geometry encoders for 3D hand motion recovery from egocentric video. The system achieves state-of-the-art results on multiple benchmarks, significantly improving accuracy in occlusion-heavy scenarios and for hands that go out of sight.

PageIndex is a Python-based document indexing system designed for vectorless, reasoning-based Retrieval-Augmented Generation (RAG). It focuses on enhancing information retrieval for AI agents.

LiquidAI has introduced LFM2.5-DSpark, a new model designed to significantly accelerate inference speeds. This advancement aims to improve the efficiency of deploying large language models in real-world applications.

Researchers have developed a novel multi-dimensional, primitive-based framework for dynamic contrast-enhanced MRI (DCE-MRI) reconstruction. This method disentangles anatomy, contrast enhancement, and motion into separate temporal basis functions, enabling geometrical interpretation and competitive performance with conventional methods.

OpenAI is reinforcing its commitment to Zero Data Retention for eligible API users. The company is also previewing a new feature called Private Safety Processing, designed to enhance AI safety without compromising user data privacy.