engblogs

TransActV2 leverages over 16,000 lifelong user actions, next action loss, and system optimizations to drastically improve Pinterest's personalized home feed recommendations at scale with ultra-low latency.

Exploring strategies and innovations to counteract malicious applications of artificial intelligence as of June 2025.

Exploring strategies implemented to address The New York Times' data requests while ensuring robust user privacy protection.

Stripe Tax offers a comprehensive, self-serve solution for global businesses to register, calculate, collect, and file taxes efficiently across multiple jurisdictions.

Explores Pinterest's automated solution, Hadoop Control Center (HCC), for seamless in-place migration and scaling of large Hadoop™ clusters on AWS, overcoming challenges of traditional manual processes with enhanced data safety and operational efficiency.

Leveraging Lambda's 1-Click Clusters with dstack simplifies distributed multi-turn reinforcement learning training of reasoning agents using the RAGEN framework for scalable and efficient AI development.

The Common Pile v0.1 delivers an 8TB openly licensed and public domain text dataset for large language model training, alongside Comma v0.1 models, promoting transparency, openness, and reproducible AI research.

Chrome has achieved a 22% performance boost on the Speedometer 3.1 benchmark through advanced rendering pipeline optimizations, saving users millions of hours annually.

Exploring the concept of malleable software to restore user agency by enabling adaptable, composable tools over rigid apps, fostering community-driven customization and smoother transitions from user to creator.

Leveraging Anthropic Claude models, AWS infrastructure, and Snorkel AI's data platform to transform pharmaceutical data analytics through scalable, accurate, and cost-effective AI-driven insights and automation.

DeepSeek-R1-0528, the open-source AI model with advanced reinforcement learning and FP8 quantization, delivers superior reasoning, code generation, and real-time object detection capabilities via Lambda's Inference API.

Explore the enhanced capabilities of Gemini 2.5 in advanced audio dialog processing and generation.