This is a cool idea, but I wonder if it's more efficient to continue using binaries....

This is a cool idea, but I wonder if it's more efficient to continue using binaries. Let the LLM work at compile time instead of runtime.