Hugging Face has released a tool that allows users to create custom CUDA kernels for their models. This is made possible through the integration of Codex and Claude, which provide the ability to generate code based on natural language prompts. Users can now write agent skills in Python, allowing them to customize model behavior without needing extensive knowledge of low-level programming languages like CUDA C. --- Why it matters: This matters because it simplifies the process of customizing AI models for specific tasks, making it more accessible to researchers and developers who may not have expertise in low-level programming. This could lead to faster development and deployment of tailored models. Source: https://huggingface.co/blog/custom-cuda-kernels-agent-skills