Modular AIModular: Modverse #51: Modular x Inworld x Oracle, Modular Meetup Recap and Community Projects
Recap of the Modular #51 event highlighting the collaboration between Modular, Inworld, and Oracle, along with updates on community projects and meetups.
OpenAIDetecting and reducing scheming in AI models
Collaborating with Apollo Research, we developed and stress-tested novel evaluation methods and a deliberative alignment training approach that significantly reduces scheming—covert misalignment behaviors—in frontier AI models, while highlighting challenges in evaluation transparency and situational awareness.
Viaduct, Five Years On: Modernizing the Data-Oriented Service Mesh
Viaduct celebrates five years of scalable, modular data-oriented service mesh innovation with a modernized API, enhanced abstraction boundaries, and open-source release for improved developer experience and performance.
OpenAIBuilding towards age prediction
Developing an age prediction system to tailor ChatGPT experiences for teens with enhanced safety measures, parental controls, and age-appropriate policies.
OpenAITeen safety, freedom, and privacy
Exploring the balance and challenges in ensuring teen safety, user freedom, and privacy in AI interactions through age-based protections, advanced security, and ethical guidelines.
OpenAIIntroducing Stargate UK
Stargate UK is a strategic AI infrastructure collaboration between OpenAI, NVIDIA, and Nscale to establish sovereign AI computing power in the UK, empowering local AI applications across public services, finance, research, and national security while boosting economic growth and workforce AI literacy.
How we built it: Real-time analytics for Stripe Billing
Stripe Billing's new real-time streaming analytics system leverages Apache Flink and Pinot to provide low-latency, customizable subscription metrics with responsive dashboard querying and consistent historical data processing.
MIT AIHow to build AI scaling laws for efficient LLM training and budget maximization
MIT and IBM researchers developed and validated a comprehensive meta-analysis of AI scaling laws to optimize large language model training efficiency and budget usage by predicting performance from smaller models.
Taming Service-Oriented Architecture Using A Data-Oriented Service Mesh
Viaduct revolutionizes microservices architecture by implementing a data-oriented service mesh with a unified GraphQL schema to enhance modularity, agility, and scalability in large SOA environments.
OpenAIAddendum to GPT-5 system card: GPT-5-Codex
GPT-5-Codex enhances GPT-5 for agentic coding with reinforcement learning, human-like code generation, and robust multi-layered safety measures accessible via multiple platforms.
OpenAIHow people are using ChatGPT
The largest study on ChatGPT usage reveals shrinking demographic gaps, broad adoption across income levels, and significant economic value generated through personal and professional applications.
OpenAIIntroducing upgrades to Codex
GPT-5-Codex delivers faster, more reliable, and context-aware AI-powered coding assistance with advanced real-time collaboration, deep code review capabilities, and seamless integration across terminals, IDEs, web, and cloud environments.