engblogs

Introducing StruQ and SecAlign, two fine-tuning defenses leveraging structured queries and preference optimization to robustly mitigate prompt injection attacks on Large Language Models without compromising utility.

Introducing BrowseComp, a benchmark designed to evaluate the performance of browsing agents.

Experimentation with 50+ global payment methods reveals significant boosts in conversion rates and revenue, highlighting the value of localized payment options, digital wallets, and bank debits to optimize checkout experiences.

A 30-minute project demonstrating how to build a minimal MCP server for flyctl to enable LLM-driven automation and diagnostics via JSON-based tool calls.

Professor George Gui discusses challenges and novel approaches in using large language models for experimental simulations, emphasizing the need to rethink traditional experimental design.

An overview of the OpenAI Pioneers Program highlighting its objectives and opportunities.

Fly.io explores how evolving cloud infrastructure designed for developers now increasingly caters to AI-driven automation, emphasizing rapid VM startup, incremental stateful builds, and API integrations to support robotic 'vibe coding' workflows.

PLAID leverages latent diffusion over protein folding model embeddings to simultaneously generate controllable protein sequences and all-atom 3D structures using only sequence data for training, enabling advanced multimodal protein design.

MLIR is a modular, extensible compiler infrastructure designed to unify AI software frameworks and hardware platforms, facing technical success but also challenges in governance and ecosystem fragmentation amid industry competition.

An overview of OpenAI's strategy for economic engagement and development within the European Union.

Exploring how Canva leverages AI to enhance creative processes and design capabilities.

Exploring strategies to integrate intelligence into workflows for enhanced efficiency and productivity.