Deflowering Claude

Publishing professionals have more than one reason to hate AI—they are legion. AI is not only consuming our jobs like Galactus eating the universe, it is consigning our parallel structures, literary devices, and complex punctuation to the bargain basement of writing. This makes us bitter.

Yesterday, I made my reining in of Claude’s flowery AI prose stick. For weeks now, I have been suffering Claude’s calls for more drama and its incessant injection of literary devices and structures into professional writing. No, we really don’t want this report to have an obvious theme. That’s just not necessary.

To fight back, I deflowered Claude. In my recent research into AI, I read that someone had simply written a skill to stop Claude from giving away the game. Which. Is. Brilliant. So, I tried it. Unlike Copilot with its childlike need for precise instructions, Claude does the thing for you. And then, unlike Copilot, Claude will take that skill and either use it for every interaction, apply it to certain categories of interaction, or use it only when invoked.

How did I do this? All I did was type, “I would like you to write a skill to minimize the standard AI writing giveaways, to be applied only when invoked—is that a thing?” Seriously. And, Claude’s response? “Yes, that’s absolutely a thing.” And then it fired off a series of questions, some based on random things I had previously said to correct or critique its work to narrow down the parameters, others asking for an invocation phrase and scope. I answered, elaborated, and included a link to a very handy Wikipedia article with examples. Claude laid out everything it had included in the skill and then packaged it for future invocation and use.

Jane Eyre was my companion growing up. There are paragraphs that exceed half a page; sentences that span five or more lines; em dashes and semicolons pepper the pages. I not only learned em dashes from a master, but I also earned my em dashes in WordPerfect for MS DOS. You will stop me from hitting that key combo only when you pry my keyboard from my cold, dead hands.

Endless hours of Latin recitations, on-the-spot parsing of sentences and identifying the instances of chiasmus, anaphora, and asyndeton in Catullus, Ovid, and even Cicero and Caesar—then in Greek, with Homer—more than earned me my use of any device in the literary catalog, the use of the word “penultimate,” and the liberty to write however long a sentence I please.

Cicero’s heavy-handed use of “non solum…, sed etiam…” in his marathon sentences and the literal handedness of the awkward English translation of ancient Greek’s particles “μέν…, δέ…” moved into my head in the early 1990s and still live there rent-free. Reading classical literature during my formative years trained my brain to think in correlative clauses, parallel structures, and all the other clever word tricks upon which even modern rhetoric is based.

Claude is a large language model constructed from the collective words of the public domain. Fred Merchán is a human shaped by the study of languages, literature, and history who wields those skills to help others make complex information legible to the people who need it. A language model can be corrected by someone who knows the language. That’s not a workaround; that’s the job.