Just Write

LLMs are trained on work that humans created. But now we’re telling humans to change how they write because “you sound like an LLM.”

When you see “It’s not x, it’s y”. Or em-dashes, or other common constructs in AI writing, those are there because millions of documents by millions of people contain those things.

I’ve worked with a lot of people who produced articles or books without having any writing background. They learned to write by emulating things they saw other people do on blogs, and then they learned to write by adopting a style guide I asked them to use.

Writers learn to write by copying things they read. It starts early in life and continues through high school and beyond. Your boss wants a certain memo format. Your company wants a certain brand voice. Buzzfeed wants clickbait headlines, and algorithms reward building things up to keep you engaged longer. LLMs swallowed up the entire Internet and are now emulating human writing. Those LLMs didn’t just train on peer-reviewed books; they trained on Reddit posts, community blogs, personal websites, and user-generated comments.

So what happens when we change our writing because people think LLMs wrote it?

The LLMs will change, too.

That’s the sad part. Readers are so distrustful of LLM-generated content that they don’t even trust the real stuff anymore.

So just write. Share what you know.

(I wrote this post by hand, by the way.)