Search: Grounded Safety Evaluation

Models & Launches

Aymara AI Launches Safety Evaluation System for 20 Language Models

CurrentLens
May 1, 2026

Aymara AI unveils a platform for custom safety evaluations of large language models, revealing performance gaps.

Latest
Trending

Agents & Automation

OpenAI Launches Three Academy Courses on Agents and Workflows

CurrentLens
Jun 13, 2026

OpenAI released three Academy courses focused on practical AI skills, building repeatable workflows, and applying agents in everyday work.

Models & Launches

Google Releases Gemini-SQL2; Gemini 3.1 Pro Scores 80.04% on BIRD

CurrentLens
Jun 13, 2026

Google Research announced Gemini-SQL2, a Gemini 3.1 Pro-powered text-to-SQL capability that posted 80.04% execution accuracy on the BIRD single-model leaderboard.

Science & Healthcare

Africa CDC and WHO launch $518M continental Ebola response plan

CurrentLens
Jun 6, 2026

A six-month 'One Response' plan targets the Bundibugyo Ebola outbreak with unified coordination, surveillance, clinical care and community engagement across affected countries.

Policy & Safety

HASC adds right-to-repair language to FY27 defense policy bill

CurrentLens
Jun 6, 2026

The House Armed Services Committee inserted right-to-repair provisions into its FY27 defense policy draft, aiming to ease barriers that limit troops' ability to fix equipment.

AI Creative

Startups Pull Users Off Phones With In-Person Games and DIY Cyberdecks

CurrentLens
Jun 6, 2026

TechCrunch highlights founders building physical social products: Board raised funding for in-person games, and cyberdeck DIYs are going viral.

Agents & Automation

MicroPython WASM Sandbox Enables Safer Datasette Plugin Execution

CurrentLens
Jun 6, 2026

Simon Willison published an alpha MicroPython-in-WASM sandbox (micropython-wasm) and a Datasette plugin (datasette-agent-micropython) to run plugin code with constrained access.

Models & Launches

DKPS method cuts model-evaluation queries using cached responses

CurrentLens
Jun 6, 2026

An arXiv paper introduces a DKPS-based approach that uses cached model outputs to predict benchmark scores while substantially reducing the number of queries.