The Archive

Dario Amodei outlines Anthropic's engagement with U.S. Department of War on national security AI applications.

In this post, we dive into one of the most critical workloads in modern AI: Flash Attention, where you’ll learn: How to implement Flash Attention using NVIDIA... In this post, we dive into one of the most critical workloads in modern AI: Flash Attention, where you’ll learn: Environment requirements: See the quickstart doc for more information on installing cuTile Python. The attention mechanism is the computational heart of transformer models. Given a sequence of tokens, attention enables each token to “look at” every other… Source

A computation is considered deterministic if multiple runs with the same input data produce the same bitwise result. While this may seem like a simple property... Source

OpenAI releases GPT-5.4, frontier model with 1M-token context, state-of-the-art coding, computer use, and tool search capabilities.

No content provided for evaluation.

OpenAI introduces CoT-Control, finding reasoning models struggle to control chains-of-thought, highlighting monitorability as a safety safeguard.

OpenAI introduces AI education tools, certifications, and measurement resources to help schools address AI capability gaps.

OpenAI outlines five sequential AI value models for business transformation from workforce fluency to process reinvention.

OpenAI launches Adoption news channel offering frameworks and insights for AI business deployment.

VfL Wolfsburg scales ChatGPT across operations by prioritizing people and organizational capability over isolated pilots.

OpenAI ships ChatGPT for Excel with GPT-5.4 and financial data integrations for modeling and regulated environments.

GPT-5.2 Pro assists in deriving and verifying graviton tree amplitudes, extending single-minus amplitudes to quantum gravity.

OpenAI introduces Learning Outcomes Measurement Suite to assess AI's impact on student learning across educational contexts.

Axios uses AI to augment local journalism workflows and reporter productivity while maintaining editorial standards at scale.

NVIDIA ACE is a suite of technologies for building AI agents for gaming. ACE provides ready-to-integrate cloud and on-device AI models for every part of in-game... NVIDIA ACE is a suite of technologies for building AI agents for gaming. ACE provides ready-to-integrate cloud and on-device AI models for every part of in-game characters, from speech to intelligence to animation. To run these models alongside the game engine efficiently, the NVIDIA In-Game Inferencing (NVIGI) SDK includes a set of performant libraries that developers can integrate into C++… Source

NVIDIA CUDA Tile is one of the most significant additions to NVIDIA CUDA programming and unlocks automatic access to tensor cores and other specialized... NVIDIA CUDA Tile is one of the most significant additions to NVIDIA CUDA programming and unlocks automatic access to tensor cores and other specialized hardware. Earlier this year, NVIDIA released cuTile for Python, giving Python developers a natural way to write high-performance GPU kernels. Now, the same programming model is available in Julia through cuTile.jl. In this blog post… Source

Google DeepMind releases Gemini 3.1 Flash-Lite, the fastest and most cost-efficient model in the Gemini 3 series.

OpenAI publishes system card documenting GPT-5.3 Instant capabilities, limitations, and safety properties.

OpenAI releases GPT-5.3 Instant, a faster variant optimized for everyday conversational use cases.

Cohere C-suite guide on enterprise AI advantages: productivity, competitive advantage, and 2026 adoption strategies.

Import AI 447 explores AGI economic models, game-based AI evaluation, and multi-agent ecosystem dynamics.

Autonomous networks are quickly becoming one of the top priorities in telecommunications. According to the latest NVIDIA State of AI in Telecommunications... Autonomous networks are quickly becoming one of the top priorities in telecommunications. According to the latest NVIDIA State of AI in Telecommunications report, 65% of operators said AI is driving network automation, and 50% named autonomous networks as the top AI use case for ROI. Yet many telcos still report gaps in AI and data science expertise. This makes it difficult to scale safe… Source

To make 6G a reality, the telecom industry must overcome a fundamental challenge: how to design, train, and validate AI-native networks that are too complex to... To make 6G a reality, the telecom industry must overcome a fundamental challenge: how to design, train, and validate AI-native networks that are too complex to be tested in the physical world. The NVIDIA Aerial Omniverse Digital Twin (AODT) solves this by enabling a continuous integration/continuous development (CI/CD)-style workflow where Radio Access Network (RAN) software is trained… Source

OpenAI signs agreement with Department of War establishing safety red lines and deployment protocols for classified AI systems.

Anthropic responds to Secretary of War Pete Hegseth's comments on AI policy and provides guidance to customers.

Alibaba has introduced the new open source Qwen3.5 series built for native multimodal agents. The first model in this series is a ~400B parameter native... Alibaba has introduced the new open source Qwen3.5 series built for native multimodal agents. The first model in this series is a ~400B parameter native vision-language model (VLM) with reasoning built with a hybrid architecture of mixture of experts (MoE) and Gated Delta Networks. Qwen3.5 can understand and navigate user interfaces, which improves on the previous generation of VLMs. Qwen3.5… Source

Organizations deploying LLMs are challenged by inference workloads with different resource requirements. A small embedding model might use only a few gigabytes... Organizations deploying LLMs are challenged by inference workloads with different resource requirements. A small embedding model might use only a few gigabytes of GPU memory, while a 70B+ parameter LLM could require multiple GPUs. This diversity often leads to low average GPU utilization, high compute costs, and unpredictable latency. The problem isn’t just about packing more workloads onto… Source