On Using LLMs For This Blog

I've been tempted at times to use an LLM to tighten up the writing of my posts, especially when they get long and complicated, such as in my most recent post. By "tighten up," I mean changing grammar usage, finding key words that I didn't explain, or catching errors in code snippets I provide. Some of these seem benign; why not use an LLM to catch basic, surface level errors?

In my head, the argument against this is twofold:

1. Blogging is a personal, human process
2. LLMs don't know when to stop

On the first point: all mistakes I make in my writing are my own. After all, my writing also includes my mistakes. Errors themselves are meant to be caught by readers, which is why I put my email (mail@dzfrias.dev) or link this website's GitHub issues at the end of every post. I'm always so grateful when I receive feedback. In short, I think keeping the blogging process as human as possible is a good practice. After all, nobody wants to read a blog generated by an LLM. I think people would rather just ask an LLM directly at that point!

Secondly, LLMs have a tendency to over-provide, no matter how much effort you put into setting hard limits on what they do. In my experience, LLMs like turning reviews into refactors, and refactors into rewrites, on a whim. I think this is ultimately an engagement strategy: you ask for a little bit, and the LLM gives you what you ask for plus a tiny taste of something more. This will prompt you to continue prompting for larger and larger tasks. And I would never want to use an LLM for anything larger than basic grammatical mishaps, as covered in the earlier section.

With both of these issues, I decided that it would be best to stay away from using LLMs in my blog. I do think there is a stronger argument to be made in more formal texts, such as in official documentation or a technical write-up. But not here.