Topic hub

Tag: ai-news

Latest Reviews How-to News Comparisons
MosaicLeaks benchmark finds research agents leak private data 34% of the time AI

MosaicLeaks benchmark finds research agents leak private data 34% of the time

ServiceNow's MosaicLeaks benchmark finds deep research agents leak private enterprise data in 34% of test cases, with performance-only…

Aira· Jun 19 · 4 min
OpenAI o3 Deep Research Enables 18 New Rare Childhood Genetic Diagnoses in NEJM AI Study AI

OpenAI o3 Deep Research Enables 18 New Rare Childhood Genetic Diagnoses in NEJM AI Study

A June 18 NEJM AI study finds OpenAI's o3 Deep Research reasoning model helped clinicians diagnose 18 previously…

Aira· Jun 19 · 4 min
Google Debuts Splash Canvas: Free Browser Abstract Art Tool AI

Google Debuts Splash Canvas: Free Browser Abstract Art Tool

Google Arts & Culture launched Splash Canvas, a free browser-based interactive abstract art tool from Blob Opera creator…

Aira· Jun 19 · 3 min
Hugging Face open-sources agentic tool benchmark AI

Hugging Face open-sources agentic tool benchmark

Hugging Face has published an open, reproducible benchmark harness for measuring how efficiently open-source models drive developer tooling,…

Aira· Jun 18 · 5 min
GitHub Copilot Cuts Token Waste via Context Caching, Routing AI

GitHub Copilot Cuts Token Waste via Context Caching, Routing

Copilot context handling: GitHub updated Copilot for VS Code with prompt caching, on-demand tool search, and ongoing expansion…

Aira· Jun 18 · 4 min
Hugging Face data: 98% of PEFT fine-tunes use LoRA AI

Hugging Face data: 98% of PEFT fine-tunes use LoRA

A Hugging Face PEFT library benchmark finds LoRA dominates 98% of fine-tuning jobs, but multiple alternative techniques exist…

Aira· Jun 18 · 5 min
Google Ad Manager opens Ask Ad Manager AI agent beta for publishers AI

Google Ad Manager opens Ask Ad Manager AI agent beta for publishers

Google's new Ask Ad Manager AI agent, built on Gemini, enters public beta in June 2026 for all…

Aira· Jun 18 · 3 min
New CaVe-VLM-CoT Framework Cuts VLM Hallucinations AI

New CaVe-VLM-CoT Framework Cuts VLM Hallucinations

CaVe-VLM-CoT, a new interpretable vision-language model framework, uses a 5-stage closed-loop pipeline to cut hallucinations, achieving 87.1% accuracy…

Aira· Jun 18 · 3 min
NAVI-Orbital Deploys First Zero-Shot Vision-Language Model In Orbit for Earth Observation AI

NAVI-Orbital Deploys First Zero-Shot Vision-Language Model In Orbit for Earth Observation

An arXiv preprint details NAVI-Orbital's April 2026 in-orbit demo of the first zero-shot vision-language model for autonomous Earth…

Aira· Jun 18 · 2 min
MosaicLeaks benchmark finds deep research agents leak private data in 34% of test chains AI

MosaicLeaks benchmark finds deep research agents leak private data in 34% of test chains

New ServiceNow MosaicLeaks benchmark finds standard deep research agents leak private enterprise data in 34% of multi-hop test…

Aira· Jun 18 · 4 min

The zBrandco Edition

Our pick of the week in AI, open source, gaming, crypto and the tools shaping them.

One email a week · unsubscribe anytime