✷ The archive · 11 labs · 3,054 dispatches
The labs
No more opening a dozen official blogs every morning. First-hand releases from OpenAI, Anthropic, DeepMind and the rest, each with its substance pulled out.
Google DeepMind AI for Math Initiative
Google DeepMind and Google.org have launched the AI for Math Initiative, a partnership with five prestigious research institutions to accelerate mathematical discovery using AI tools like Gemini Deep Think, AlphaEvolve, and AlphaProof.
The Shifting Global Compute Landscape: China's Rise in AI Hardware and Software
Hugging Face analyzes how U.S. export controls have paradoxically accelerated China's development of domestic AI chips and compute-efficient open-weight models, creating a parallel AI infrastructure.
NVIDIA Isaac for Healthcare v0.4: Building a Healthcare Robot from Simulation to Deployment
NVIDIA Isaac for Healthcare v0.4 introduces an end-to-end workflow for building autonomous surgical assistants using the SO-ARM starter workflow, combining synthetic simulation data with real-world training.
OpenAI gpt-oss-safeguard Technical Report
OpenAI has released gpt-oss-safeguard-120b and gpt-oss-safeguard-20b, open-weight reasoning models designed specifically for content classification based on provided policies.
gpt-oss-safeguard release notes / what's new
OpenAI has released gpt-oss-safeguard, a set of open-weight reasoning models designed for safety classification tasks that allow developers to apply custom policies at inference time.
Anthropic Research: Emergent Introspective Awareness in Large Language Models
Anthropic research indicates that Claude models, particularly Opus 4 and 4.1, demonstrate a limited but genuine capacity to monitor and control their own internal neural states.
Anthropic Expands to Japan with Tokyo Office and AI Safety Partnership
Anthropic has opened its first Asia-Pacific office in Tokyo and signed a Memorandum of Cooperation with the Japan AI Safety Institute to establish international AI evaluation standards.
OpenAI gpt-oss-safeguard models released via Ollama
OpenAI and Ollama released the open‑source gpt-oss-safeguard safety‑reasoning models (20B and 120B) under Apache 2.0, enabling organizations to run policy‑driven content classification with transparent reasoning.
NVIDIA Isaac for Healthcare v0.4 Release
NVIDIA Isaac for Healthcare v0.4 introduces a developer framework and the SO-ARM starter workflow to enable end-to-end AI healthcare robotics development from simulation to real-world deployment.
Dai Nippon Printing (DNP) ChatGPT Enterprise Deployment Results
Dai Nippon Printing (DNP) achieved a 100% weekly active usage rate and a 95% reduction in patent research time after deploying ChatGPT Enterprise across ten core departments.
IBM Granite 4.0 Nano Release Notes
IBM has released Granite 4.0 Nano, a series of high-performance small language models (SLMs) designed for edge and on-device applications under an Apache 2.0 license.
Doppel AI Defense System: Automating Threat Detection with GPT-5 and o4-mini
Doppel has deployed an AI-driven defense system using OpenAI's GPT-5 and o4-mini models to automate the detection and takedown of impersonation sites and phishing attacks, reducing response times from hours to minutes.
OpenAI Corporate Recapitalization and the Launch of the OpenAI Foundation
OpenAI has completed a corporate recapitalization, establishing the OpenAI Foundation as a nonprofit that controls the for-profit OpenAI Group PBC to ensure AGI benefits all of humanity.
Microsoft and OpenAI Partnership Update 2025
Microsoft and OpenAI have signed a new definitive agreement that transitions OpenAI to a public benefit corporation and updates IP rights, AGI verification, and compute agreements.
Hugging Face Voice Cloning with Consent
Hugging Face has introduced the voice consent gate, a system infrastructure that requires speakers to explicitly state their consent via a recorded phrase before a voice cloning model can be activated.
MiniMax M2 Release on Ollama Cloud
MiniMax M2 is a high-performance model optimized for coding and agentic workflows, now available on Ollama's cloud platform.
OpenAI Strategy for US AI Leadership and Energy Infrastructure
OpenAI advocates for a national project to build 100 gigawatts of new energy capacity annually to maintain US leadership in AI and avoid an electron gap with China.
GPT-5 System Card Addendum: Sensitive Conversations
OpenAI has updated GPT-5 to better recognize and support users in emotional distress, reducing suboptimal responses by 65-80% through collaboration with over 170 mental health experts.
OpenAI Strengthens ChatGPT Responses for Sensitive Conversations
OpenAI has updated ChatGPT's default model (GPT-5) to better recognize and support users in distress, reducing undesired responses in mental health, self-harm, and emotional reliance scenarios by 65% to 80%.
huggingface_hub v1.0: Five Years of Building the Foundation of Open Machine Learning
huggingface_hub has reached v1.0 after five years, providing core access to over 2 million models, 500k datasets, and 1M Spaces while introducing breaking changes like httpx, hf_xet, and a redesigned CLI to support the next decade of open machine learning.
Hugging Face Streaming Datasets Update
Hugging Face has optimized the streaming capabilities of the datasets library, reducing startup requests by 100x and increasing data resolution speed by 10x to enable efficient training on multi-terabyte datasets without local downloads.
Steuerrecht.com Case Study: Scaling Legal Analysis with ChatGPT Business
Steuerrecht.com uses ChatGPT Business to automate routine legal drafting, research, and multilingual communication, reducing certain tasks from days to minutes and saving up to 10 hours of work per week per lawyer.
Advancing Claude for Financial Services
Anthropic has expanded Claude for Financial Services with a new Excel add-in, real-time market data connectors, and specialized Agent Skills for complex financial modeling and analysis.
T5Gemma: Encoder-Decoder LLMs via Decoder-Only Adaptation
Google DeepMind introduces T5Gemma, a collection of encoder-decoder models created by adapting pretrained decoder-only Gemma 2 models to improve inference efficiency and reasoning performance.
MedGemma and MedSigLIP Release Notes
Google DeepMind has released MedGemma and MedSigLIP, a collection of open multimodal models designed to provide a high-performance, privacy-preserving foundation for healthcare and lifesciences AI development.
Gemma 3n Release Notes / What's New
Google DeepMind has released Gemma 3n, a mobile-first multimodal model family featuring a nested MatFormer architecture that allows for elastic inference and high performance on edge devices.
Gemini 2.5 Flash-Lite Release Notes
Google DeepMind has released the stable version of Gemini 2.5 Flash-Lite, the fastest and most cost-efficient model in the Gemini 2.5 family, featuring a 1 million-token context window and native reasoning capabilities.
Google DeepMind Veo: Integrating Generative Video into Live-Action Filmmaking for ANCESTRA
Google DeepMind partnered with director Eliza McNitt and Darren Aronofsky's Primordial Soup to produce the short film ANCESTRA, demonstrating how Veo can blend generative video with live-action footage through personalized generation and motion matching.
AlphaEarth Foundations: Unified Geospatial Embeddings for Global Mapping
Google DeepMind has introduced AlphaEarth Foundations, an AI model that integrates petabytes of multimodal Earth observation data into unified digital embeddings to revolutionize global mapping and environmental monitoring.
Mistral AI Studio: A Production Platform for Enterprise AI
Mistral AI has launched AI Studio, a production-grade platform designed to move enterprise AI from prototypes to reliable, governed systems through three core pillars: Observability, Agent Runtime, and AI Registry.
DeepMind Backstory experimental tool for image provenance and trustworthiness
Google DeepMind announced Backstory, an experimental AI tool that uses Gemini models to reveal an image's AI‑generation status, provenance, alterations, and evolving context, helping users assess trustworthiness.
Gemini Deep Think achieves gold-medal standard at IMO 2025
An advanced version of Gemini with Deep Think solved five out of six International Mathematical Olympiad problems, earning 35 points and achieving a gold-medal standard.
Google DeepMind Aeneas: AI for Contextualizing Ancient Inscriptions
Google DeepMind has introduced Aeneas, the first AI model designed to help historians interpret, attribute, and restore fragmentary Latin inscriptions through multimodal analysis and parallel text retrieval.
Genie 3 real-time world model release
Genie 3 is DeepMind’s new real‑time world model that generates interactive 720p environments at 24 fps and maintains consistency for minutes, enabling advanced embodied‑agent research.
Google DeepMind Perch Update: Advancing Bioacoustics for Endangered Species
Google DeepMind has released an updated open-source Perch model that improves bird species prediction and expands bioacoustic analysis to mammals, amphibians, and underwater environments like coral reefs.
Google DeepMind Deep Loop Shaping for LIGO Gravitational-Wave Observatories
Google DeepMind introduced Deep Loop Shaping, an AI method that reduces control noise in LIGO's feedback systems by 30 to 100 times, enabling more precise detection of gravitational waves.
Gemini 2.5 Deep Think ICPC World Finals Performance
An advanced version of Gemini 2.5 Deep Think achieved gold-medal level performance at the 2025 International Collegiate Programming Contest (ICPC) World Finals, solving 10 out of 12 complex algorithmic problems.
Google DeepMind Discovers New Fluid Dynamics Singularities Using AI
Google DeepMind has developed a novel AI method using Physics-Informed Neural Networks (PINNs) to discover new families of unstable singularities in fluid dynamics equations, providing a path toward solving long-standing mathematical challenges.
LeRobot v0.4.0 release notes / what's new
Hugging Face has released LeRobot v0.4.0, introducing scalable Datasets v3.0, VLA models like PI0.5 and GR00T N1.5, and a plugin system for streamlined hardware integration.
Google DeepMind Frontier Safety Framework Update
Google DeepMind has released the third iteration of its Frontier Safety Framework (FSF), introducing new risk domains for harmful manipulation and misaligned AI, and adding Tracked Capability Levels (TCLs) for earlier risk detection.
Gemini Robotics 1.5 and Gemini Robotics-ER 1.5 Release
Google DeepMind has introduced Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, a pair of models that combine high-level embodied reasoning with direct vision-language-action control to enable robots to perform complex, multi-step physical tasks.
Google DeepMind CodeMender AI Agent for Code Security
Google DeepMind has introduced CodeMender, an AI-powered agent that automatically identifies, patches, and proactively secures software vulnerabilities in open-source projects.
Google DeepMind and Commonwealth Fusion Systems Partnership for Fusion Energy
Google DeepMind is partnering with Commonwealth Fusion Systems to apply AI, reinforcement learning, and the TORAX simulator to accelerate the achievement of net fusion energy in the SPARC tokamak.
Gemini 2.5 Deep Think Release
Google DeepMind has released Deep Think in the Gemini app for Google AI Ultra subscribers, introducing parallel thinking techniques to enhance complex problem-solving in coding, science, and mathematics.
Google DeepMind and Kaggle Launch Game Arena for AI Evaluation
Google DeepMind and Kaggle have introduced Game Arena, an open-source platform that evaluates AI models through head-to-head competition in strategic games to provide a verifiable measure of general problem-solving intelligence.
OpenAI Acquires Software Applications Incorporated to Integrate Sky into ChatGPT
OpenAI has acquired Software Applications Incorporated, the creators of Sky, to integrate deep macOS integration and natural language interfaces into ChatGPT.
Consensus Scholar Agent: Research Automation with GPT-5 and Responses API
Consensus has launched Scholar Agent, a multi-agent system powered by GPT-5 and the Responses API that automates the synthesis of peer-reviewed literature across 220 million papers.
Hugging Face and Meta introduce OpenEnv for agentic environments
Hugging Face and Meta have launched OpenEnv and the OpenEnv Hub, a standardized, open community hub for secure, sandboxed agentic environments used in training and deployment.
OpenAI Introduces Company Knowledge for ChatGPT Business, Enterprise, and Edu
OpenAI launched Company Knowledge for ChatGPT Business, Enterprise, and Edu, letting users query internal tools like Slack and Google Drive via a GPT‑5 model with citations while keeping data private.
OpenAI Economic Blueprint for South Korea
OpenAI has released an Economic Blueprint for South Korea, proposing a dual-track strategy to combine sovereign AI capabilities with strategic frontier AI partnerships to drive national productivity.