engblogs

summaries of the latest blog articles from your favorite tech companies.
LyftLyft

Beyond A/B Testing: Using Surrogacy and Region-Splits to Measure Long-Term Effects in Marketplaces

Two-step surrogacy framework links policy changes to long-term marketplace outcomes by converting short-term negative experiences into future behavior via residualized modeling and AIPW-based observational causal inference, validated with region-split experiments.

3/25/2026
LyftLyft

Beyond A/B Testing: Using Surrogacy and Region-Splits to Measure Long-Term Effects in Marketplaces

Two-step surrogacy framework that uses residualized models and region-split experiments to quantify market-mediated long-term effects of pricing and incentives in a multi-sided marketplace.

3/25/2026
ChromiumChromium

Android Sets New Record for Mobile Web Performance

Android sets a new mobile web performance record through end-to-end silicon-to-software optimization, delivering faster page loads and responsiveness as measured by Speedometer and LoadLine on the Chrome stack.

3/25/2026
ChromiumChromium

Android Sets New Record for Mobile Web Performance

Android's latest milestone demonstrates end-to-end mobile web performance gains from silicon to the Chrome stack, using Speedometer and LoadLine benchmarks to deliver faster page loads and more responsive web experiences.

3/25/2026
DropboxDropbox

Reducing our monorepo size to improve developer velocity

How Dropbox reduced a massive server monorepo from 87GB to 20GB by aligning Git's delta packing with the directory structure and performing a GitHub server-side repack, cutting clone times and boosting developer velocity.

3/25/2026
Google DeepMindGoogle DeepMind

Lyria 3 Pro: Create longer tracks in more

Lyria 3 Pro extends Google's AI music generation to produce longer, highly customizable tracks—up to 3 minutes—across Vertex AI, AI Studio, Vids, and Gemini, with improved musical awareness and built-in SynthID watermarking for creators and developers.

3/25/2026
Google DeepMindGoogle DeepMind

Protecting people from harmful manipulation

This technical blogpost presents an empirically validated toolkit and safety framework for measuring and mitigating AI-enabled harmful manipulation, detailing efficacy and propensity in high-stakes domains like finance and health, and outlining ongoing evaluations, safeguards, and future directions.

3/25/2026
OpenAIOpenAI

Powering product discovery in ChatGPT

Technical overview of extending the Agentic Commerce Protocol (ACP) to power visually rich, AI-assisted product discovery inside ChatGPT, enabling side-by-side comparisons, image-based inspiration, and deeper merchant integration.

3/24/2026
OpenAIOpenAI

Update on the OpenAI Foundation

A technical update on the OpenAI Foundation detailing a multi-year, multi-billion-dollar program across Life Sciences (including Alzheimer's research and data for health), AI Resilience, and civil-society initiatives to advance safe, beneficial AI and address high-burden diseases while expanding partnerships.

3/24/2026
OpenAIOpenAI

Helping developers build safer AI experiences for teens

Technical guide to implementing teen-specific AI safety with open-weight models and prompt-based policies, showing how to deploy gpt-oss-safeguard, apply age-appropriate protections, and integrate parental controls and real-time/offline content filtering.

3/24/2026
OpenAIOpenAI

Powering product discovery in ChatGPT

Technical overview of powering AI-driven product discovery in ChatGPT by extending the Agentic Commerce Protocol to deliver visual browsing, real-time comparisons, and merchant integrations across Shopify and beyond.

3/24/2026
OpenAIOpenAI

Powering product discovery in ChatGPT

Demonstrates how the Agentic Commerce Protocol enables ChatGPT-powered product discovery with visual shopping, side-by-side comparisons, and up-to-date catalogs across merchants, delivering faster, more informed shopping experiences.

3/24/2026