XNews

OpenAI AI models automate new model training including GPU kernels

According to The Information, OpenAI's internal AI models can now automate the training pipeline for new experimental models, including writing GPU kernel code and optimizing training code. Researchers only need to provide an optimization example, and the AI takes weeks to implement and test similar improvements. Multiple internal AI agents now collaborate autonomously to solve problems, reducing experiments that previously took years to about one week.

Source posts