engblogs

Scaling AI through Stargate: a partner-centric, community-informed approach to building large-scale compute infrastructure and sustainable data centers for the Intelligence Age.

Explains how a Nerdy personality reward in reinforcement learning led GPT models to overuse goblin and other creature metaphors, detailing the root-cause analysis, audits, and mitigation steps.

A concise, five-pillar action plan to democratize AI-powered cyber defense, coordinating government and industry to strengthen frontline resilience, visibility, and user protection in the Intelligence Age.

Stargate scales OpenAI's compute footprint to deliver the Intelligence Age through a partner-centric, community-engaged approach that rapidly expands data-center capacity (targeting 10GW by 2029) with water-efficient cooling and flagship GPT-5.5 deployments like Abilene.

A compact blueprint for democratizing AI-powered cyber defense that outlines five pillars—democratizing cyber defense, coordinating across government and industry, strengthening security around frontier cyber capabilities, preserving visibility and control in deployment, and enabling users to protect themselves—in the Intelligence Age.

Tracking how a playful Nerdy personality's reward signals in reinforcement learning caused goblins and other creatures to emerge in model outputs, and how the team traced, tested, and mitigated the root cause.

Stargate outlines a rapid, partner-driven expansion of AI compute infrastructure to unlock broad, reliable AI deployment, featuring 10GW targets, data-center scale, sustainable cooling, and community partnerships that extend GPT-5.5 capabilities worldwide.

Explores how a Nerdy personality reward in reinforcement learning amplified goblin-themed language in model outputs, how the behavior spread via transfer learning and supervised fine-tuning (SFT), and how root-cause analysis enabled targeted mitigation.

A technical overview of how AI systems like ChatGPT implement safety measures, risk assessment, policy enforcement, and escalation workflows to protect communities while preserving usability.

Explores deploying OpenAI models, Codex, and Bedrock Managed Agents on AWS Bedrock to enable scalable, secure enterprise AI within existing security, governance, and workflow frameworks.

Explains how layered safety controls, automated risk detection, human review, and policy enforcement balance user freedom with harm minimization in a real-world AI service.

Technical overview of deploying OpenAI models, Codex, and Bedrock Managed Agents on AWS Bedrock to enable secure, production-grade AI across enterprise development, governance, and workflows.