AirbnbMigrating Airbnb’s JVM Monorepo to Bazel
Airbnb migrated their extensive JVM monorepo from Gradle to Bazel over 4.5 years, significantly improving build speed, reliability, and developer experience through automated build file generation, remote build execution, and sophisticated dependency management.
OpenAIScaling accounting capacity with OpenAI
Basis leverages advanced OpenAI models, including GPT-5, to build AI agents that automate and enhance accounting workflows, enabling firms to save time and expand advisory capacity through improved reasoning, task delegation, and transparency.
OpenAIOpenAI’s letter to Governor Newsom on harmonized regulation
OpenAI urges Governor Newsom to lead harmonized AI regulations aligning state, federal, and global standards to foster innovation, ensure safety, and maintain US leadership in democratic AI development.
EleutherAIPretraining Data Filtering for Open-Weight AI Safety
A comprehensive study demonstrating that scalable, multi-stage pretraining data filtering effectively reduces dangerous knowledge in open-weight large language models with minimal performance tradeoffs and tamper-resistant safeguards.
Fly.ioGames as Model Eval: 1-Click Deploy AI Town on Fly.io
Exploring gamified AI model evaluation through a one-click deploy of AI Town on Fly.io, highlighting dynamic behavioral insights beyond traditional benchmarks.
Fly.ioThe Future Isn't Model Agnostic
AI product success depends on deeply understanding and specializing a single model rather than pursuing model agnosticism amid converging benchmark performances.
OpenAIHow Cursor uses GPT-5
An overview of how Cursor integrates and leverages the advanced capabilities of GPT-5 to enhance its functionalities.
OpenAIIntroducing GPT-5
GPT-5 is OpenAI's most advanced AI model, delivering state-of-the-art performance across coding, writing, health, and multimodal understanding with improved reasoning, safety, and reliability features for diverse real-world applications.
OpenAIHow Amgen uses GPT-5
Amgen leverages GPT-5 to enhance research, safety, and business solutions through advanced AI integration.
OpenAIGPT-5 System Card
GPT-5 introduces a multi-model system with a real-time router optimizing model selection for faster, more accurate responses, enhanced safety, and improved performance in writing, coding, and health domains.
OpenAIFrom hard refusals to safe-completions: toward output-centric safety training
GPT-5 introduces safe-completion training, a technique that enhances AI safety by focusing on output safety and maximizing helpfulness, particularly improving handling of dual-use prompts compared to refusal-based approaches.
OpenAIFirst look at GPT-5
An initial overview highlighting the key features and advancements of GPT-5 based on OpenAI's latest research update.