Thomas Ptacek bans LLM phrasing and its praise from drafts

Thomas Ptacek laid out two rules for editing with a language model in a September 17 post: never adopt its words, never accept its praise.

The stakes are about who's actually doing the writing. Ptacek says readers can spot a model's phrasing "in the parts per trillion," because everything it writes reads like a magazine headline.

Rule one keeps any phrase the model wrote out of the finished draft. Rule two is about judgment. The praise a model gives a first draft reinforces weak choices. It cuts off the rethinking that produces a distinct voice.

What he keeps the model for is narrower. Catching repeated words. Flagging modifiers that don't earn their place. Pointing out a paragraph sitting in the wrong order. A copyeditor, not a ghostwriter.

Martin Fowler raised a different kind of objection the same day. Fowler's objection grants that LLMs are useful enough that avoiding them may be irresponsible. Then he argues he dislikes them anyway, for confidently bullshitting. When corrected, he says, they offer fake remorse instead of an actual correction.

Ptacek is setting a workflow rule for someone who already uses the tool daily. Fowler is naming a values objection that survives even after granting the tool works.