engblogs

An empirical analysis on finding optimal model size and training tokens for a given compute budget reveals that current large language models are too large for their compute budget and lack sufficient training data.

Discovering systematic errors with cross-modal embeddings - This blog post introduces Domino, a new approach for discovering systematic errors made by machine learning models. It discusses a framework for quantitatively evaluating methods like Domino and explores the task of slice discovery in machine learning models.

A blogpost providing updates and guidance on Privacy Sandbox testing for the ads relevance and measurement proposals, including information on origin trials, API testing, feedback channels, and updated settings and controls.

The blogpost discusses the latest Chrome beta channel release for Android, Chrome OS, Linux, macOS, and Windows, and covers topics including reducing user agent string information, origin trials, federated credential management API, audio output latency, font-palette and custom palettes, HWB CSS function, window.open() changes, MediaCapabilities API for WebRTC, Secure Payment Confirmation API V3, USBDevice forget() method, and WebUSB sameObject behavior. It also mentions the deprecation and removal of WebSQL in third-party contexts.

Uniting pen & paper with software to create dynamic models.

The blogpost discusses the challenges of grading interactive coding assignments and proposes a method that uses reinforcement learning to grade assignments based on gameplay rather than code analysis.

GopherCite is a model that aims to address the problem of language model hallucination by supporting answers with verified quotes.

A blogpost describing GopherCite, a model that tackles language model hallucination by providing verified quotes as evidence for its factual claims.

Exploring the latest advancements in GPT-3 with enhanced editing and inserting features.

The blogpost discusses how Chrome became the highest scoring browser on Apple's Speedometer 2.0 benchmark, detailing the improvements and optimizations made to achieve this milestone in browser performance.

Using the Strangler pattern to migrate a monolithic codebase to a fully-fledged BFF for the public API

Collaboration between AI and historians to restore, place, and date ancient texts using Ithaca, a deep neural network for textual restoration