engblogs

OpenAI and PwC collaborate to embed AI agents across core finance workflows—planning, forecasting, procurement, and close—enabling automated, governance-driven CFO processes with human oversight at scale.

Domain Adaptive Learning for Carrot Ads transfers knowledge from Instacart’s data-rich marketplace to new partner domains, addressing cold-start with efficient, first-party data-driven CTR modeling.

A concise technical update on Modular's latest developments, covering AMD AI DevDay highlights, new Modular offices, and a thriving community driving continuous shipping of AI models.

Explores Olmix, a framework for offline data mixing in language model development, detailing how proxy models, a log-linear regression surrogate, and optimization define robust domain mixtures and how mixture reuse enables efficient updates as data evolves.

A practical framework for data mixing in large-language-model development, pairing an offline mixing schema with a mixture-reuse approach to adapt to evolving data and optimize performance with lower recomputation.

Model-signature-driven, 'Send What You Use' feature trimming in Pinterest's root-leaf ML serving reduces unnecessary data sent from root to leaf, cutting bandwidth, latency, and cost while enabling scalable multi-model deployments.

Feature Trimmer uses model signatures to send only the features each leaf model needs, eliminating unused data in the root-leaf path to cut network bandwidth, improve latency, and slash infrastructure costs in Pinterest's ML serving.

Streamlined opt-in Advanced Account Security enhances ChatGPT and Codex protection with phishing-resistant sign-in via passkeys or hardware security keys, stricter account recovery, shorter sessions, and user-visible activity alerts.

A technical overview of OpenAI's Advanced Account Security, an opt-in suite that strengthens ChatGPT and Codex account protection with phishing-resistant sign-in, fortified recovery, shorter sessions, and privacy-preserving training options for high-risk users.

Introducing Advanced Account Security: an opt-in, phishing-resistant security framework for ChatGPT and Codex that tightens sign-in, recovery, and session management for high-risk users and defenders.

Explores the AI co-clinician initiative to augment clinicians in healthcare by enabling triadic, multimodal, telemedicine-enabled care with clinical safeguards and evidence-grounded decision support.

A technical roadmap to democratize AI-powered cyber defense in the Intelligence Age, outlining five pillars to reinforce cybersecurity, coordinate across sectors, secure frontier cyber capabilities, preserve deployment visibility, and empower users to protect themselves.