I wouldn't be surprised if we're already at the point of more LLM-written Rust than hand-written. Models training off models