All published AI news and analysis.

OpenAI's Astra model is the first to achieve the Critical cybersecurity capability threshold under the Preparedness Framework. This designation highlights enhanced safeguards implemented for its release.

A new study systematically compares LLM-as-a-judge (LLMaJ) methods for Natural Language Generation (NLG) evaluation, both with and without task decomposition. The research found no evidence that decomposing evaluation tasks improves performance over a non-decomposed baseline. Instead, reported gains in previous decomposition-based LLMaJ methods appear to stem from the use of human labels as training data.

Researchers have introduced StainPresetNet, a novel framework for stain normalization in pathological images that combines structural preservation with dataset-level color mapping. This method offers computational efficiency and multi-directional adaptability without the need for retraining.

OpenAI has released version 0.152.0 of its Codex CLI. This update introduces new features and improvements to the command-line interface for interacting with Codex models.

OpenAI's ChatGPT Ads has achieved a $1 billion annualized revenue run rate and is expanding its availability worldwide. This move aims to increase access to AI technologies through both free and cost-effective offerings.

Polimill is developing Japan's next-generation public AI infrastructure by utilizing OpenAI's GPT models and Codex. This initiative aims to enhance how municipalities access and utilize administrative knowledge, thereby speeding up development processes.

OpenAI has partnered with Thailand's Ministry of Higher Education, Science, Research and Innovation (MHESI) to launch an eight-week accelerator program. This initiative aims to support 10 startups in the health, wellness, and education sectors.

Researchers Maayan Sharon and Tom Hope introduced RATIO, a new benchmark for retrieving scientific literature. RATIO focuses on three specific 'ideation moves': Address, Broaden, and Specify, to better support scientific inspiration and hypothesis generation for both humans and AI.

Researchers have identified Visual Retrieval Heads (VRHs), a small subset of attention heads (1.7-2.6%) in Vision-Language Models (VLMs) that are causally responsible for grounding text descriptions to specific image regions. These VRHs generalize across various visual reference tasks and are architecturally shared across different VLM architectures.

OpenAI has released version 0.151.0 of its Codex CLI. This update is part of ongoing development for the command-line interface tool.