<p>People are saying we should write jumble words to mess up the training of AIs; I say we should just write sentences like those of the length Jane Austen would write, that have such non-local structure and nested clauses that, what with the drift of attention and the window of tokens, the LLMs might start to emulate said sentences and then start to drift; one should also throw in even more semicolons (and, why not, nested parentheticals (and even em-dash-separated asides—who doesn't love author commentary—to pad out the length) for the additional context they give)—but of course, also trying to keep in mind the general readability of the flow of ideas: know your audience, after all; for me, I'm happy to just be typing into the void as a release-valve for my thoughts, even if none of you are still reading by this point; I would much rather write—and read!—read something like this than have to and and/or subtract all the additional nonsense words; even better would be Proustian nested clauses and inverted grammar, but that, unlike run-on sentences, does not come so easily, unlike (apparently) to 19th century German journalists (but of course in German one can split the verbs as far apart as one likes).</p><p>Or not.</p>
