engblogs

Building the infrastructure for delivering labor market insights from LinkedIn data to create economic opportunity for every member of the global workforce.

Injecting vision into frozen speech models for zero-shot audiovisual automatic speech recognition (AV-ASR), achieving state-of-the-art performance on AV-ASR benchmarks while maintaining decent performance on traditional audio-only benchmarks.

Performing semantic search with Cohere and OpenSearch, including step-by-step instructions to spin up a local OpenSearch instance, store Cohere embeddings in an index, and retrieve and use these embeddings.

Exploring a novel informal ink augmentation approach that uses weak recognition and tagging to enhance drawing tools by preserving natural pen-and-paper interaction while enabling implicit grouping and selection affordances.

Exploring the OpenAI Cybersecurity Grant Program for cutting-edge research and innovation in the field.

A blogpost showcasing the world's fastest unified matrix multiplication using the Modular framework.

A real-time system to catch anti-patterns in Spark applications at Uber scale, helping developers optimize their apps.

Facilitating development of AI-powered cybersecurity capabilities for defenders through grants and support.

This blogpost discusses the methods used to ensure a successful launch of Ads on Netflix, including testing the system, technologies involved, and best practices developed.

A visual-language model called REVEAL, which utilizes a multi-source multi-modal "memory" to answer knowledge-intensive queries, overcoming the limitations of existing retrieval-augmented models and reducing the need for massive amounts of training data and parameter size.

Highlights of the new features and updates in Git 2.41

Enhancing mathematical cognition through guided processes