It's Dangerous to Go Alone. Take Books.

You need to be reading books. Specifically, books by humans. Not for self-improvement or to look sophisticated. As a survival tactic. I am not about to launch into a Robin Williams routine from Dead Poets Society about the riches of the inner life, though I reserve the right to do so at any time in the future. This is more urgent than that, more pragmatic. The world is turning into a lunatic gauntlet of illusion and deceit, and if you’re not reading books, you’re not gonna make it.

Yesterday, the U.K. outlet Press Gazette reported that at least six publications, including Wired and Business, have had to take down multiple articles that apparently were written by AI. A “journalist” called Margaux Blanchard had been successfully pitching human interest stories about tech and business until, after a series of increasingly sketchy emails, editors came to suspect she was a bot.

The funniest one was a piece in Wired called “They Fell in Love Playing Minecraft. Then the Game Became Their Wedding Venue,” about two people who definitely exist and definitely were married after they met playing the online world-building game. No they didn’t, no they don’t, no they weren’t. Fake article by a fake writer about a fake wedding in a fake world. Run at a site dedicated to sharp analysis of digital technology.

The real scandal, though, was buried about halfway through an article on the matter in The Guardian. Wired, admitting its mistake, said: “This story did not go through a proper fact-check process or get a top edit from a more senior editor.” Or it did, and now said senior editor is looking for some intern’s scalp to offer up. I wish I was even a little bit surprised. But online publications are constantly churning through more content than they can vet properly to feed the ravenous maw of the internet.

The point is, the scammer operating “Margaux” succeeded because there was no trained human with enough spare attention to consider his (and its) submissions discerningly. Budget pressures being what they are in journalism, this will happen again. There are going to be made-up articles out there trying to trick you. Which means you, the reader, are going to need to tune up your clanker radar (your clank-dar? It’s a work in progress. That’s one way you know I’m not a machine).

There is only one way to get better at telling bot slop apart from human prose, and that is by reading a lot of human prose. I know this because I am a professional and lifetime reader of human prose, and I am very good at distinguishing between AI-generated content from the real thing. I just have a feel for it. By comparison, I’m a total amateur when it comes to the visual arts, and I’m way more easily fooled by image generators. You learn to know what you love.

Share

So when Large Language Models started overproducing em-dashes, I wasn’t worried—though it’s probably my favorite punctuation mark. I can’t tell you in precise terms how I know this, but the way Joan Didion uses em-dashes is fundamentally different from the way ChatGPT does. By design, the machine’s style is more rote. This is true even when it tries to imitate the greats. It can only do so in the most average way possible, since taking averages is literally what it’s designed to do—what it has to do to work.

I wouldn’t even rule out the possibility that an LLM could help me identify some of the basic trends I’m noticing in certain cases. But then, to know if it was right, I’d have to rely on—you guessed it—my own taste and discernment. Plus I’m talking more broadly here not just about one or two principles, but an entire sensibility comprised of limitless intuitions, judgment cases, and unconscious subtleties. The em-dashes are just one example. It’s a skill I learned over a lifetime.

And you can too. You need to. The best way to do it is to read books. Online articles won’t do. Most of them are already composed of just the same anodyne banalities that were then fed into our LLMs to train them. AI tends to write like the people who were already trying to optimize their content for the algorithm to begin with—that is, people who were already writing for machines.

Instead, you should be looking for books that smell like must and feel like fresh thought. You should be reading the weirdest books you can find, pulling the threads of your niche interests until you end up in some rural library or second-hand bookstore, run by a man in overalls with the beard of a portal mage, thumbing through the out-of-print memoir of a minor Burmese functionary from the British empire. That’s the skill you need to compete in today’s media environment. It’s dangerous to go alone. Take books.