Interesting article, I wonder how much longer new language models will be detectable like this.