> ## Content Index
> Fetch the complete content index at: https://birchtree.me/llms.txt
> Use this file to discover other available public pages before exploring further.

# What is a perversion to writing?
- URL: https://birchtree.me/blog/what-is-a-perversion-to-writing/
- Published: 2026-08-19T13:00:00.000Z
- Updated: 2026-08-19T13:00:00.000Z
- Author: Matt Birchler

Simon Wilson has the best description of LLMs that I’ve seen so far. He called them [a calculator for words](https://simonwillison.net/2023/Apr/2/calculator-for-words/?ref=birchtree.me), and I think it’s a great way to put it.

A traditional calculator takes numbers and operators as input, does some math, and outputs an answer. LLMs are essentially the same thing, except the input and output are words. You input some words, it does some math on them, and it gives you an answer. They’re tools that do a job. We can debate the implications of this sort of calculator existing, but I think it’s a really useful way to describe them.

I bring this up because the definition of writing, and what constitutes a "perversion of writing", has been…let's just say…a topic of conversation this week. This debate revolves around watermarking text that AI generates. The complaint here is that of the enormous amount of math being done to generate the text models deliver already, adding a specific bit more is an affront to everything.

*Quick aside,* [*this video*](https://youtu.be/FnAqruxx-QE?ref=birchtree.me) *does a good job explaining how watermarking systems work in LLMs if you want to actually understand what's being discussed.*

As far as what I think about watermarking text so that software can determine whether it was generated by AI, **fuck yeah**. I think anything generated with AI, whether it’s images, video, code, or text, should be able to be checked by the user if they want to know.

Arguments have been made about how text is somehow different from images or video, and I think that's incredibly misguided, and shows a lack of respect for other mediums. Sure, writing can be sacred, but nothing these language models generate is sacred. Like number calculators, these word calculators are hugely useful, but no one is going to type something complex into a TI-83 Plus, gasp, and go, "hang it in the fucking Louvre".

The general population has clearly indicated that they want to know when something is AI-generated. I think it is a particularly nihilistic view of the world to suggest that people do not deserve to know this if they want to find out.