✷ The archive · 11 labs · 3,415 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.
OpenAI Codex Mobile Integration and Enterprise Updates
OpenAI has integrated Codex into the ChatGPT mobile app, allowing users to manage long-running agentic tasks across laptops and remote environments from their phones.
VeRL-Omni: RL Training Framework for Diffusion and Omni-Modality Models
vLLM has announced the pre-release of VeRL-Omni, a general reinforcement learning post-training framework designed specifically for multimodal generative models, including diffusion and omni-modality architectures.
Unlocking asynchronicity in continuous batching
Hugging Face shows how to overlap CPU and GPU work in continuous batching using CUDA streams and events, cutting LLM inference time by about 22% for an 8B model generating 8K tokens.
vLLM Elastic Expert Parallelism
vLLM introduces Elastic Expert Parallelism (Elastic EP), enabling Mixture-of-Experts (MoE) deployments to scale the number of workers up or down at runtime without requiring a server restart.
OpenAI Updates ChatGPT Safety to Improve Context Recognition in Sensitive Conversations
OpenAI has implemented safety updates to ChatGPT that enable the model to better recognize evolving risk signals and harmful intent across single and multiple conversations, specifically for suicide, self-harm, and harm-to-others scenarios.
Anthropic 2028 AI Leadership scenarios
Anthropic released a paper outlining two possible 2028 outcomes for US‑China AI competition, arguing that maintaining a compute lead for democracies is crucial to prevent authoritarian AI dominance.
Anthropic and Gates Foundation Partnership for Global AI Impact
Anthropic is partnering with the Gates Foundation to provide $200 million in grant funding, Claude usage credits, and technical support to advance global health, life sciences, education, and economic mobility over four years.
Anthropic and PwC Expand Strategic Alliance for Enterprise AI Deployment
Anthropic and PwC have expanded their strategic partnership to deploy Claude, Claude Code, and Claude Cowork across PwC's global workforce and client engagements to automate complex enterprise functions.
Building a safe, effective sandbox to enable Codex on Windows
OpenAI has implemented a custom sandbox for Codex on Windows to provide secure, restricted execution of coding agents without requiring constant user approval for every command.
OpenAI Response to TanStack npm Supply Chain Attack
OpenAI responded to a supply chain attack involving the TanStack npm library as part of the Mini Shai-Hulud campaign, rotating code-signing certificates for macOS, iOS, Windows, and Android apps without evidence of user data compromise.
How finance teams use ChatGPT Work – OpenAI Academy guide
OpenAI’s Academy article shows how finance teams can use ChatGPT Work to turn existing financial inputs into review-ready assets for reporting, planning, and variance analysis without writing code.
Google DeepMind Co-Scientist: A Multi-Agent AI System for Scientific Hypothesis Generation
Google DeepMind has introduced Co-Scientist, a multi-agent AI partner powered by Gemini that iteratively generates, debates, and evolves novel scientific hypotheses to accelerate research in life sciences and engineering.
How NVIDIA uses OpenAI Codex with GPT-5.5
NVIDIA engineers and researchers are utilizing Codex, powered by GPT-5.5 and running on GB200 and GB300 infrastructure, to automate complex engineering tasks and end-to-end machine learning research workflows.
OpenAI Parameter Golf: Insights on AI-Assisted ML Research
OpenAI's Parameter Golf challenge revealed how AI coding agents lower experimentation barriers and surface exceptional ML talent through tightly constrained optimization problems.
AutoScout24 Scales Engineering with OpenAI AI-Powered Workflows
AutoScout24 Group has integrated ChatGPT and Codex to reduce development timelines from weeks to days and increase engineering throughput across its vehicle marketplace platform.
Building Blocks for Foundation Model Training and Inference on AWS
Hugging Face details a four-layer architectural framework on AWS—comprising infrastructure, resource orchestration, ML software stacks, and observability—to support the evolving scaling laws of pre-training, post-training, and test-time compute.
ChatGPT Adoption Trends Q1 2026
OpenAI reports that in Q1 2026, ChatGPT usage broadened across age groups, gender, and geography, with a shift toward more specialized workplace tasks.
OpenAI Launches Campus Network Interest Form for Student Clubs
OpenAI announced the Campus Network, inviting student clubs worldwide to join via an interest form to receive AI learning support, event resources, early tool access, and ambassador opportunities.
How Enterprises Scale AI: Lessons from OpenAI's Executive Research
OpenAI identifies five key patterns for scaling AI in enterprises, emphasizing that success depends more on culture, governance, and workflow redesign than on the technical rollout of tools.
OpenAI launches OpenAI Deployment Company to accelerate enterprise AI adoption
OpenAI has launched the OpenAI Deployment Company, a standalone business unit backed by $4 billion in initial investment and the acquisition of Tomoro, designed to embed Forward Deployed Engineers into organizations to redesign workflows around frontier AI.
vLLM Performance Benchmarks: Topping Artificial Analysis Leaderboard
vLLM has achieved top rankings on the Artificial Analysis leaderboard for DeepSeek V3.2, MiniMax-M2.5, and Qwen 3.5 397B by implementing aggressive kernel fusion and speculative decoding optimizations.
vLLM TurboQuant Study: Accuracy and Performance Analysis
A comprehensive study by vLLM reveals that FP8 remains the superior default for KV-cache quantization, while TurboQuant variants trade significant throughput and latency for additional memory capacity.
Running Codex safely at OpenAI
OpenAI has implemented a security framework for Codex that combines sandboxing, managed network policies, and agent-native telemetry to balance developer productivity with enterprise-grade control.
Adaptive Parallel Reasoning: The Next Paradigm in Efficient Inference Scaling
BAIR introduces Adaptive Parallel Reasoning (APR), a paradigm that allows LLMs to dynamically decide when to parallelize reasoning threads to reduce latency and avoid context-rot while maintaining accuracy.
Anthropic Research: Teaching Claude Why to Reduce Agentic Misalignment
Anthropic has reduced agentic misalignment in Claude models by shifting from training on aligned behaviors to teaching the underlying ethical principles and reasoning behind those behaviors.
OpenAI GPT-5.5 and GPT-5.5-Cyber Release
OpenAI has introduced GPT-5.5-Cyber in limited preview and the Trusted Access for Cyber (TAC) framework to provide verified security defenders with more permissive, specialized AI capabilities for critical infrastructure protection.
Parloa AI Agent Management Platform (AMP) Overview
Parloa has launched the AI Agent Management Platform (AMP), an enterprise-grade system built on OpenAI models including GPT-5.4, designed to allow non-technical subject matter experts to design and deploy reliable, low-latency customer service agents.
OpenAI introduces GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper in the API
OpenAI launched three new realtime audio models—GPT‑Realtime‑2, GPT‑Realtime‑Translate, and GPT‑Realtime‑Whisper—to enable developers to build voice apps that reason, translate, and transcribe live.
Simplex Software Development Integration of OpenAI Codex
Simplex has integrated OpenAI Codex and ChatGPT Enterprise to shift from assistive AI to AI-native delivery, achieving up to 70% reduction in screen development time for CRUD-based web applications.
Trusted Contact in ChatGPT: OpenAI's New Safety Feature for Crisis Support
OpenAI introduced Trusted Contact, an optional safety feature in ChatGPT that lets adults nominate a trusted person to be notified if the system detects serious self‑harm concerns.
Anthropic Natural Language Autoencoders
Anthropic has introduced Natural Language Autoencoders (NLAs), a method that converts internal model activations into human-readable text to reveal a model's hidden thoughts and motivations.
Anthropic Donates Petri Alignment Tool to Meridian Labs
Anthropic has released Petri 3.0 and transferred ownership of the open-source alignment toolbox to the nonprofit Meridian Labs to ensure independent and neutral AI model evaluation.
The Anthropic Institute Research Agenda
Anthropic has launched The Anthropic Institute (TAI) to study AI's real-world impacts on economics, security, and society using internal frontier lab data to inform public policy and safety.
vLLM V1 Migration: Ensuring Backend Correctness in Reinforcement Learning
ServiceNow AI achieved parity between vLLM V0 and V1 for RL rollout generation by fixing logprob semantics, runtime defaults, weight update paths, and implementing an fp32 lm_head.
AlphaEvolve: Scaling Gemini-Powered Algorithmic Discovery Across Industries
Google DeepMind's AlphaEvolve, a Gemini-powered coding agent, has demonstrated significant impact by optimizing algorithms in genomics, quantum physics, AI infrastructure, and commercial sectors.
How ChatGPT Learns and Protects User Privacy
OpenAI outlines the data sources, privacy-preserving technologies like the OpenAI Privacy Filter, and user controls available to prevent ChatGPT conversations from being used in model training.
Serving Agentic Workloads at Scale with vLLM x Mooncake
vLLM integrates Mooncake's distributed KV cache store to boost agentic LLM serving, delivering 3.8× higher throughput, 46× lower TTFT, and 8.6× lower end‑to‑end latency on realistic traces while scaling to 60 GB200 GPUs.
Hugging Face Open ASR Leaderboard: Private Datasets to Combat Benchmaxxing
Hugging Face has introduced private evaluation datasets from Appen Inc. and DataoceanAI to the Open ASR Leaderboard to prevent test-set contamination and provide a more robust measure of real-world ASR performance.
OpenAI B2B Signals: How Frontier Firms are Pulling Ahead
OpenAI introduces B2B Signals to reveal that frontier firms—those in the 95th percentile of usage—now use 3.5x more intelligence per worker than typical firms, driven by complex, agentic workflows.
OpenAI Introduces ChatGPT Futures: Class of 2026
OpenAI has launched ChatGPT Futures, a program providing grants and model access to students in the Class of 2026 who use AI to build tools, advance research, and create social impact.
Singular Bank Integrates ChatGPT and Codex to Launch Singularity Assistant
Singular Bank has developed Singularity, an internal AI assistant powered by ChatGPT and Codex, reducing banker productivity losses by 60 to 90 minutes per day through automated portfolio analysis and communication.
Uber and OpenAI Integration: AI-Powered Driver Guidance and Voice Booking
Uber is integrating OpenAI frontier models to launch Uber Assistant for driver earnings optimization and voice-driven ride booking to reduce friction for riders.
Grok Imagine Quality Mode API Release
xAI has launched Quality Mode for the Grok Imagine API, providing enterprise developers with enhanced realism, improved multilingual text rendering, and superior creative control for image generation and editing.
xAI Grok Connectors Release
xAI has launched Connectors, deep integrations for Grok on web, iOS, and Android that allow the AI to read and write data across third-party applications like Google Workspace, GitHub, and Notion.
xAI and Anthropic Compute Partnership
xAI has partnered with Anthropic to provide access to the Colossus 1 supercomputer, with additional plans to explore orbital AI compute capacity.
Anthropic Expands Compute Capacity via SpaceX Partnership and Increases Claude Usage Limits
Anthropic has partnered with SpaceX to access the Colossus 1 data center, enabling higher usage limits for Claude Code and the Claude API.
GPT-5.5 Instant System Card
OpenAI has released the GPT-5.5 Instant system card, marking the first time an Instant model is classified as High capability in Cybersecurity and Biological & Chemical Preparedness categories.
GPT-5.5 Instant release notes / what's new
OpenAI has released GPT-5.5 Instant, a new default model for ChatGPT that improves factuality, reduces verbosity, and enhances personalization through deeper integration with user context.
OpenAI MRC Multipath Reliable Connection Protocol
OpenAI has released the Multipath Reliable Connection (MRC) protocol via the Open Compute Project to improve GPU networking resilience and performance in large-scale AI training clusters.
OpenAI European Youth Safety Blueprint and EMEA Youth & Wellbeing Grant
OpenAI announced its European Youth Safety Blueprint and the first recipients of the EMEA Youth & Wellbeing Grant to promote age‑appropriate AI use and protect young people in Europe, the Middle East, and Africa.