engblogs

Explores establishing a global, safety-first framework for youth in AI—advocating an international AI Safety Institute, age-appropriate protections, risk-based governance, parental controls, and transparent accountability to unlock safe learning and opportunity.

Explains how to build, deploy, and share focused internal sites and lightweight apps with Codex for team workflows like launch trackers, dashboards, and onboarding pages.

Codex expands across roles, tools, and workflows with new plugins, interactive Sites, and in-place annotations that tailor Codex to business, design, sales, data analytics, and investing workflows, enabling live dashboards, hosted apps, and an open plugin ecosystem for enterprise teams.

Travelers scales an AI-powered auto claims assistant nationwide by integrating OpenAI's real-time API and frontier models into claims infrastructure to automate first notice of loss, improve customer experience, and efficiently handle catastrophe surges.

A technical guide showing how Codex enables ChatGPT Sites to quickly build, deploy, and share lightweight internal sites and apps—such as launch trackers, dashboards, and onboarding pages—with hosting, access controls, and simple automation.

A technical overview of Travelers' nationwide deployment of an AI Claim Assistant powered by OpenAI, detailing real-time model integration with claims infrastructure to automate auto property loss filings at scale and sustain 24/7 service during catastrophe surges.

Airbnb demonstrates how sequential geographic recovery signals and a hierarchical Bayesian framework using informative priors enable corridor-level forecasting when local data is scarce, effectively turning geography into a time machine for learning across markets.

Sequential geographic recovery signals power a hierarchical Bayesian forecasting approach that propagates informative priors across corridors, enabling reliable corridor-level demand forecasts when local data is scarce.

Instacart's Semantic IDs paper describes compressing product embeddings with a residual vector quantizer and contrastive learning over a hierarchical catalog to create scalable, cross-category codes that improve cold-start, discovery, and catalog health.

Rebuilding Instacart's ads retrieval as a generative, SID-based autoregressive model that spells out product concepts to significantly expand catalog coverage and discoverability, while boosting relevance and brand diversity via beam-search generation and a GPU-backed serving stack.

Technical breakdown of a PRC-linked influence campaign that used AI-generated social content to portray US data centers and AI infrastructure as driving higher electricity costs, detailing the operational workflow, platform-evasion tactics, and cross-platform manipulation targeting US and overseas Chinese audiences.

Analyzes a PRC-origin influence operation targeting US tech policy via coordinated inauthentic X accounts and AI-generated content, illustrating detection and policy implications.