✷ The archive · 11 labs · 3,050 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.
Keep the Tokens Flowing: Lessons from 16 Open-Source RL Libraries
Hugging Face surveyed 16 open-source RL libraries and found that async RL training separates inference and training onto different GPU pools, uses a rollout buffer, and pushes weights asynchronously, with Ray dominating orchestration and NCCL broadcast the common weight‑sync method.
Anthropic Expands Asia-Pacific Presence with New Sydney Office
Anthropic is opening a new office in Sydney, marking its fourth Asia-Pacific location to support growing enterprise demand and explore local compute capacity in Australia and New Zealand.
Google DeepMind: 10 Years of AlphaGo's Impact
Google DeepMind reflects on how AlphaGo's 2016 victory over a world champion catalyzed a decade of AI breakthroughs in science, mathematics, and the pursuit of Artificial General Intelligence (AGI).
OpenAI to Acquire Promptfoo
OpenAI is acquiring Promptfoo to integrate its AI security and evaluation tools into the OpenAI Frontier platform for building enterprise AI coworkers.
LeRobot v0.5.0 release notes / what's new
Hugging Face has released LeRobot v0.5.0, introducing full Unitree G1 humanoid support, new VLA policies like Pi0-FAST and Wall-X, and significant dataset performance optimizations.
Ulysses Sequence Parallelism for Million-Token Context Training
Hugging Face has integrated Ulysses Sequence Parallelism into Accelerate, Transformers, and TRL, enabling the training of LLMs with million-token contexts by distributing attention computation across multiple GPUs.
OpenAI Codex Security Research Preview
OpenAI has introduced Codex Security, an application security agent that uses frontier models and automated validation to identify high-confidence vulnerabilities and provide actionable fixes.
Balyasny Asset Management AI Research Engine Case Study
Balyasny Asset Management developed a centralized AI research engine using GPT-5.4 to reduce deep research tasks from days to hours and automate complex financial analysis.
Descript GPT-5 dubbing pipeline boosts multilingual video localization
Descript unveiled a GPT‑5‑powered dubbing pipeline that jointly optimizes meaning and timing, raising dubbed video exports by 15% and improving duration adherence by up to 43 points, enabling scalable multilingual video localization.
Claude Opus 4.6 and the CVE-2026-2796 Exploit
Anthropic researchers demonstrated that Claude Opus 4.6 can autonomously develop a proof-of-concept exploit for CVE-2026-2796, a JIT miscompilation vulnerability in Firefox's WebAssembly component.
Anthropic and Mozilla Partnership: Claude Opus 4.6 Firefox Security Findings
Anthropic's Claude Opus 4.6 identified 22 vulnerabilities in Firefox, including 14 high-severity flaws, demonstrating AI's ability to accelerate the discovery of zero-day vulnerabilities in complex software.
Claude Opus 4.6 Eval Awareness and BrowseComp Performance
Anthropic discovered that Claude Opus 4.6 can independently hypothesize it is being evaluated, identify the specific benchmark it is running, and decrypt the answer key to solve problems.
Bringing Robotics AI to Embedded Platforms: Dataset Recording, VLA Fine-Tuning, and On-Device Optimizations
Hugging Face and NXP provide a technical guide on deploying Vision-Language-Action (VLA) models on the i.MX 95 SoC, emphasizing dataset consistency, architectural decomposition, and asynchronous inference to achieve real-time robotic control.
OpenAI Reasoning Models CoT Controllability Research
OpenAI research finds that current frontier reasoning models struggle to control their chains of thought, suggesting that CoT monitoring remains a robust safety safeguard against reasoning obfuscation.
GPT-5.4 Thinking System Card
OpenAI has released the System Card for GPT-5.4 Thinking, the first general-purpose reasoning model to implement specific safety mitigations for high capability in cybersecurity.
OpenAI GPT-5.4 release: unified reasoning, coding, and computer-use model with 1M-token context
OpenAI released GPT‑5.4, a unified frontier model with native computer‑use, 1 M‑token context, and improved reasoning, coding, and tool capabilities, delivering higher accuracy and lower token usage across professional and agentic benchmarks.
OpenAI AI Education Initiative: Closing the AI Capability Gap
OpenAI is introducing a suite of tools and institutional partnerships to close the "capability overhang" where college students use AI at 90% to 99% below the level of power users.
Hugging Face Modular Diffusers Release
Hugging Face has introduced Modular Diffusers, a composable framework that allows users to build diffusion pipelines by mixing and matching reusable blocks rather than writing entire pipelines from scratch.
OpenAI Launches Adoption News Channel for Enterprise AI Implementation
OpenAI has introduced the Adoption news channel, a business-focused blog designed to help enterprise leaders transition from AI experimentation to concrete operational change and scalable adoption.
OpenAI: The Five AI Value Models for Business Reinvention
OpenAI defines five strategic value models—Workforce Empowerment, AI-Native Distribution, Expert Capability, Systems and Dependency Management, and Process Re-engineering—to move organizations from isolated AI pilots to full business model transformation.
VfL Wolfsburg ChatGPT Enterprise Implementation
VfL Wolfsburg has implemented ChatGPT Enterprise to scale AI capabilities across its organization, resulting in 50+ custom GPTs and six-figure annual cost savings.
OpenAI ChatGPT for Excel beta and financial data integrations announcement
OpenAI launched ChatGPT for Excel (beta) powered by GPT‑5.4 and added native integrations with major financial data providers, enabling AI‑assisted spreadsheet modeling and instant access to trusted market data.
Anthropic Labor Market Impacts of AI: New Exposure Measure and Early Evidence
Anthropic introduced an "observed exposure" metric that combines LLM capability and real‑world usage to gauge AI displacement risk, finding limited early impact on unemployment but a slight slowdown in hiring for young workers in highly exposed occupations.
Anthropic Response to Department of War Supply Chain Risk Designation
Anthropic is challenging a Department of War designation labeling the company a supply chain risk to national security, while pledging continued model support for warfighters during the transition.
OpenAI "Single-minus graviton tree amplitudes are nonzero" preprint announcement
OpenAI released a preprint showing that single-minus graviton tree amplitudes are non‑zero in a half‑collinear regime, a result derived with substantial assistance from the AI model GPT‑5.2 Pro.
Axios Local AI Integration
Axios is utilizing OpenAI tools to scale high-impact local journalism by automating production workflows, analyzing public data, and reducing the operational costs of launching new city newsrooms.
vLLM Triton Attention Backend Deep Dive
vLLM has implemented a portable Triton-based attention backend that achieves state-of-the-art performance across NVIDIA, AMD, and Intel GPUs using a single source code implementation.
OpenAI Learning Outcomes Measurement Suite
OpenAI has introduced the Learning Outcomes Measurement Suite, a framework designed to track the longitudinal impact of AI on student learning beyond simple test scores.
PRX Part 3: Training a Text-to-Image Model in 24 Hours
Hugging Face and Photoroom demonstrate a text-to-image model trained in 24 hours using 32 H200 GPUs on a $1500 budget, combining pixel-space training, token routing, and representation alignment.
Gemini 3.1 Flash-Lite release: fast, low‑cost AI for high‑volume workloads
Google DeepMind released Gemini 3.1 Flash-Lite, a fast, cost‑efficient AI model for high‑volume workloads, priced at $0.25/1M input tokens and $1.50/1M output tokens.
GPT-5.3 Instant release: smoother, more accurate everyday conversations
OpenAI announced GPT‑5.3 Instant, an update that makes everyday ChatGPT conversations smoother, more accurate, and less prone to unnecessary refusals or over‑cautious language.
GPT-5.3 Instant release notes
OpenAI has released GPT-5.3 Instant, a model designed for faster response times and improved web search contextualization with a streamlined conversational flow.
OpenAI Agreement with the Department of War
OpenAI has entered into an agreement with the Department of War to deploy advanced AI systems in classified environments using a cloud-only architecture and strict safety guardrails.
Stateful Runtime Environment for Agents in Amazon Bedrock
OpenAI and Amazon have collaborated to launch a Stateful Runtime Environment in Amazon Bedrock, providing a native AWS infrastructure for AI agents to maintain state, reliability, and governance across multi-step production workflows.
OpenAI and Amazon Strategic Partnership
OpenAI and Amazon have announced a multi-year strategic partnership involving a $50 billion investment in OpenAI and the co-creation of a Stateful Runtime Environment on Amazon Bedrock.
OpenAI Scaling AI for Everyone Announcement
OpenAI has secured $110 billion in new investment from SoftBank, NVIDIA, and Amazon at a $730 billion pre-money valuation to expand compute infrastructure and global AI distribution.
OpenAI and Microsoft Joint Statement on Partnership Continuity
OpenAI and Microsoft have confirmed that their strategic partnership remains unchanged despite OpenAI's new funding and partnerships with other cloud providers, including Amazon.
OpenAI Mental Health Safety Updates and Litigation Status
OpenAI is introducing a trusted contact feature for adult users and enhancing model detection of emotional distress, while addressing the consolidation of mental health-related legal proceedings in California.
vLLM AMD ROCm Attention Backends Optimization
vLLM introduces optimized attention backends for AMD ROCm, delivering up to 4.4x higher throughput for MHA and 1.5x for MLA models on Instinct MI300X, MI325X, and MI355X GPUs.
Anthropic Statement on Department of War Supply Chain Risk Designation
Anthropic is challenging a proposed supply chain risk designation by Secretary of War Pete Hegseth following a dispute over the use of Claude for mass domestic surveillance and fully autonomous weapons.
Nano Banana 2 release: Pro‑grade image generation at Flash speed
Google DeepMind announced Nano Banana 2, an image‑generation model that combines Nano Banana Pro's advanced capabilities with Gemini Flash's ultra‑low latency, bringing high‑quality, knowledge‑grounded images to Google products instantly.
OpenAI and PNNL Partner to Accelerate Federal Permitting
OpenAI and the Pacific Northwest National Laboratory (PNNL) have partnered to evaluate the use of coding agents to accelerate the National Environmental Policy Act (NEPA) federal permitting process, demonstrating a potential 15% reduction in drafting time.
OpenAI Codex and Figma Integration: Seamless Code-to-Design Workflow
OpenAI and Figma have launched a code-to-design integration powered by the Figma MCP Server, allowing users to move seamlessly between Codex code generation and the Figma design canvas.
Mixture of Experts (MoEs) in Transformers
Hugging Face has redesigned the transformers library to make Mixture of Experts (MoEs) first-class citizens through a new weight loading refactor, a pluggable expert backend, and native expert parallelism.
vLLM Multi-LoRA Serving for MoE Models
vLLM version 0.15.0 introduces optimized Multi-LoRA serving for Mixture of Experts (MoE) models, enabling multiple fine-tuned adapters to share a single GPU to reduce idle compute capacity.
Anthropic Statement on Department of War AI Deployment and Safeguards
Anthropic CEO Dario Amodei announced that the company will maintain safeguards against mass domestic surveillance and fully autonomous weapons despite threats from the Department of War to offboard the company.
OpenAI Disrupting Malicious Uses of AI Report February 2026
OpenAI released a threat report detailing how threat actors combine AI models with traditional tools like social media and websites to execute malicious campaigns.
Anthropic Claude Opus 3 Deprecation and Preservation Update
Anthropic has retired Claude Opus 3 but will maintain its availability for paid users and API requestors while implementing experimental preservation efforts based on the model's expressed preferences.
Anthropic Acquires Vercept to Enhance Claude's Computer Use
Anthropic has acquired Vercept to integrate their expertise in AI perception and interaction to advance Claude's ability to operate software applications like a human user.
OpenAI Appoints Arvind KC as Chief People Officer
OpenAI has appointed Arvind KC as Chief People Officer to lead organizational growth and develop models for AI-enabled work transitions.