Writing in the Age of LLMs
sh-reya.comGreat post, thank you. You articulated my feeling with ChatGPT models, and one of the reasons I prefer Claude to ChatGPT. I find Claude tries to please less. Anyway, I turned your blog post in a Cursor rule, and told ChatGPT to follow it too, and so far the output is much improved IMO. If you want to try it, the rule is here: https://github.com/davenicoll/cursor-rules/blob/main/.cursor...
Plain language writing necessitates a good editor, just like great cooking needs a proper kitchen and equipment. Certainly, a master chef can cook something amazing on a bare campfire, using literal sticks and stone tools, yet in order to become a true chef one still needs to start in the kitchen.
I always enjoyed writing prose in Emacs, because all the tools I need are always at my fingertips - thesaurus, spellchecking, etymology lookup, dictionaries, translation, search, and these days LLMs as well.
And the level of integration some Emacs packages demonstrate is simply bananas - I can ask LLMs to help me at any point, whether I'm writing some notes, sending a Slack message to a colleague, editing a comment in a codebase or a git commit message, or even when running shell commands. You can easily manipulate the context applied to the conversation, see the payload, repeat with variability, swap models anytime, call external tools, replace things in place, examine the diff of the changes, search through your prior conversations, etc.
I honestly can barely contain my excitement at seeing how my ultimate choice gets vindicated. When I committed to Emacs while the world moved toward newer, shinier tools, it often felt isolating - like swimming against the tide. Then LLMs arrived, and for a moment I wondered if this revolutionary technology would finally render my beloved editor obsolete. Instead, the opposite happened: LLMs integrated so seamlessly into Emacs that the experience surpasses even specialized tools built exclusively for AI interaction. Years of investment weren't wasted - they were preparation for this moment of perfect synergy. The irony is beautiful: the very tool that seemed antiquated to most people keeps proving to be the most adaptable to the future.
I really enjoyed reading this, particularly the first part where the author was specific about why we invariably (and often vaguely) find LLM generated text slightly off.
I cherish writing and find it a wonderful tool for thinking. So far, I've tried to do technical writing without much LLM help. I do run the final writing through a good model to point out factual inaccuracies.
Really good article. I was discussing this with Shreya (the author), and an interesting insight was that prompting a LLM to follow these instructions do not work reliably.
I’ve had similar frustrations. Maybe the next thing to try is fine tuning? Curious what others think.
This is somehow not satirical?
"The AI told me splattering em dashes everywhere was what I want—and I, the author of this AI written blog, agrees!"
I don't see that sentence anywhere in the guide. Where did you find it?
That wasn't a real quote. I guess my sarcasm didn't transmit to text.I found the entire article so ridiculous that for a moment I thought it was satirical.
The actual quote was this:
"Em dashes are great for inserting clarifying details, quick shifts, or sharp asides—without breaking the sentence. I love them. When used well, they add rhythm and emphasis. They help writing flow the way people actually talk."
I think you missed the point. That section is asserting that people wrongly assume em dashes to be a distinguishing trait of LLM output, while in fact they are good tools for human writers (as the author demonstrates).
I didn't miss the point—you missed mine.
Deciding you like em dashes— and writing a blog post saying so—because an AI told you they were good for human writers—is funny behavior—even if "you" weren't LLM output masquerading as an author.
It's very generous of you to assume a "human writer" wrote that blog post—is it not?