TODAY’S LINEUP

Three stories worth your time.

View archive
01 / Physical AI and robotics

Skild’s S1 lets industrial robots learn a new task from one video

NVIDIA says Skild AI’s S1 foundation model can interpret a recorded task and map it to a robot’s actions without task-specific retraining. The company describes work on unfamiliar multi-step tasks, simulation and factory deployment with NVIDIA’s physical-AI stack.

02 / Enterprise AI assistants

Amazon Quick becomes generally available on macOS and Windows

AWS has made the Amazon Quick desktop application generally available for macOS and Windows. AWS says the work assistant connects signals from email, calendar, CRM and messaging, retains audit trails, and is designed to keep enterprise data in the customer’s AWS environment.

03 / LLM inference

SageMaker adds prefix-aware routing to keep shared LLM context warm

AWS introduced prefix-aware routing for SageMaker Inference, a method that routes requests with the same prompt prefix to the same instance so its KV cache can be reused. AWS reports lower time-to-first-token and higher cache-hit rates in its Llama 3.1 70B benchmarks.

04 / Multimodal search

Bedrock Knowledge Bases adds Marengo 3.0 for video and image search

AWS says TwelveLabs Marengo Embed 3.0 is generally available in Amazon Bedrock Knowledge Bases. The integration lets teams index video, image and audio material for natural-language semantic search through a managed knowledge-base workflow.

05 / AI-assisted software development

GitHub shows how its Copilot app combines diff, terminal and browser review

GitHub published a walkthrough of using the Copilot app’s diff, terminal and browser together when working with agent-generated code. The workflow keeps code changes, command output and a running preview close enough to inspect before accepting a result.

06 / Consumer health AI

Apple adds Apple Intelligence features to health and fitness products

Apple announced new Apple Intelligence capabilities for health and fitness features, including tools intended to surface information and support personalized activity workflows. The company’s announcement also frames the release around device-level health data and user controls.

07 / Autonomous vehicles

NVIDIA outlines its physical-AI platform work with robotaxi developers

NVIDIA describes how robotaxi companies are using its computing, simulation and software stack to develop autonomous-driving systems. The post emphasizes a full development loop spanning data, training, simulation and vehicle deployment.

08 / Open models and infrastructure

AWS publishes a Qwen3.8 deployment path for SageMaker HyperPod

AWS published a technical guide for deploying Qwen3.8-2.4T-A95B with vLLM on SageMaker HyperPod. The post focuses on serving a very large mixture-of-experts model in a managed training and inference environment.

09 / AI service capacity

OpenAI pauses new Pro sign-ups while it adds capacity for Astra

OpenAI said new ChatGPT Pro subscriptions are temporarily paused because demand for Astra is placing unusual pressure on capacity. Existing Pro accounts, other plans and the API remain available, according to the company statement reported by TechCrunch.

10 / AI work automation

Amazon Quick adds a mobile activity feed for work signals

Alongside desktop availability, AWS added a mobile activity feed to Amazon Quick that consolidates email, calendar, CRM and messaging updates. AWS says agents can handle routine items while the feed prioritizes decisions that still require a person.