162 Comments
User's avatar
Cynthia Farrell's avatar

While I appreciate the guidance here, it also makes me sad. For those of us who love to write, AI has sucked the joy out of writing. Words and sentence structure and punctuation that might be part of our natural writing style is now labeled as “AI” because, well, AI is trained on real human writing. I was just reading a post on LinkedIn by someone who wrote a post on her own, in her own style, and was accused of using AI. Nearly 300 comments followed sharing similar experiences. So the net of this is that people who actually can write without AI are now being told their writing is AI and have to tie themselves in knots to not … sound like AI?

For the record, the emdash is a powerful form of punctuation that any skilled writer learned to use. AI can pry it from my hands when I’m its slave in the mines.

Ruben Hassid's avatar

its a shame - i loved em dashes, and now we cant possibly use them without being labelled as AI written

unfortunately, we have to adapt the way we write as humans

we have to be more human than AI is trying to become human

Cynthia Farrell's avatar

Sorry, but I say bullshit. Next it’s semicolons and commas and periods and then any punctuation and then conjunctions and verbs and nouns and and and … and then hell why write at all? Oops, are those ellipses an AI signal now? And if we aren’t willing or able to apply critical thinking to ascertain what is authentic, and have turned over our perceptive abilities to machines… we are just lazy. THAT is how we lose our humanity. And that is far scarier to me than someone thinking my writing is AI because I used an em dash.

Natalie's avatar

Oo yes, this right here, "And if we aren’t willing or able to apply critical thinking to ascertain what is authentic, and have turned over our perceptive abilities to machines… we are just lazy. THAT is how we lose our humanity. And that is far scarier to me than someone thinking my writing is AI because I used an em dash."

I intentionally wrote a sentence today with 3 em dashes in it. Is that normal for me? No. Do I use em dashes regularly? Yes, and I have for decades.

Out of curiosity, I ran the piece of content with the sentence through 3 different AI tools. Two of them told me that the final em dash created "too much stylistic drama." But when I prompted back with the rationale, they were full of praise (cue the eyerolling).

Cynthia Farrell's avatar

I love all of this. ❤️

Woe To The Conquered's avatar

I’m with you. As one example, I used em dashes long before AI was a thing. Ain’t gonna stop because I’m afraid someone will think it’s AI. That’s stupid.

Linda Hansen🦀🌊's avatar

AMEN!!!!! And, what about personal memories that are included? Or personal anecdotes? There's NO way AI could possibly include those in writing, but yet... (oops. I'm AI - AGAIN)

Yogita Singh's avatar

So true! Although I am not a writer in its true sense, but I am not an AI slop either. Was recently accused of using the word 'mostly' in one of my LinkedIn posts recently by a senior history professor who just wanted to argue with me on the language I used instead of the actual argument I made. Crazy times!

Jinora's avatar
2dEdited

The part about writing I have always loved is the creativity and the challenge to change syntax to meet your audience where they are. We can still do this. Creative writers and people who love their words are needed even more in the ai future 🙇🏼‍♀️

Jill Button's avatar

Was thinking the exact same thing. I’ve been writing long before AI was a thing and use many of the words that are now attributed to AI. I agree with many things like em dashes, nobody really uses the dreaded em dashes.

Yesss's avatar

Some of that style guide is useful for humans too :)

Ruben Hassid's avatar

interesting - which ones?

Yesss's avatar

The "dead AI words" were consultancy-speak before they were AI words.

Ruben Hassid's avatar

AI was most likely trained on that

Davida Oranekwu's avatar

First of all, thank you. I've been looking for how to get rid of that negative parallelism for a while now.

Secondly, you're awesome. Keep it up Ruben.

Ruben Hassid's avatar

you are welcome - happy to help :)

have fun with the file!!

Lemore's Sprint Retro's avatar

I just wrote a humoristic post about this yesterday, your lists would have saved me some time :). The Hebrew version is published English coming soon.

Ruben Hassid's avatar

did you publish it here?

Lemore's Sprint Retro's avatar

Yes - https://sprintretro.substack.com/p/5d1

If you don't read Hebrew I can send you the English version on Monday.

Monica Marcela Tamas's avatar

"That’s why I’ve put together this guide for you. In this guide, you’ll learn how to [write with pen and paper and sound like an AI]."👏wow... a lot to think about...

🤔Why did the man start handwriting his grocery list in "AI-style"?

Because he wanted to make sure his trip to the supermarket wasn't just a chore—it was a transformative journey toward optimizing his refrigerator’s ecosystem.🫪

Jean Paul Lareng's avatar

Je n'ecris qu'en francais , est ce que telcharger ces fichiers en anglais est un handicap ? Dois je les traduire en français ? Merci pour la reponse

Ruben Hassid's avatar

Je ne connais pas les AI writing typiques en francais malheureusement

Daniel Thiele's avatar

Claude pense et travaille en anglais ou en code. Tu peux appliquer les règles à n'importe quelle langue. J'ai essayé avec l'allemand et ça marche.

Prends simplement un texte standard de Claude et donne-lui ensuite le prompt suivant : "analyse le texte en fonction des modèles de documents ici et identifie et marque tous les modèles trouvés dans le texte. Crée une suggestion de modification pour chaque paragraphe identifié."

Biense's avatar

Great advice! I actually came up your profile because claude pointed me to you. I found out that the LLM itself actually knows the "I am a helpfull assistent" phrase is doing a lot of the damage you are trying to get rid of

Ruben Hassid's avatar

its just conditioning - it leans outputs toward a specific persona: polite, user-pleasing, safety-first, often sycophantic

you can add enough context on top to push against it, the behavior would change :) just like the anti-AI writing file

Robin Good's avatar

Excellent advice and info helping those who want to sound more like humans while using AI.

Question: What is not clear to me is whether you use this folder and files setup for any project / task you need to carry out, pointing always at the same Cowork main folder, or if you create separate Cowork Project folders for each project you have (and if so, where are these folders created? Under the main one?).

What do you suggest Ruben?

Ruben Hassid's avatar

1/ yes i use that Claude Cowork folder in all my tasks

2/ here is my guide on how to use an existing folder on your computer for Cowork Project: https://ruben.substack.com/i/192227474/option-3-use-an-existing-folder-on-your-computer

Suchitra's avatar

I’m curious. If everyone is using AI irrespective, just to have a skill that sounds less AI but still uses AI, isn’t a tad counterintuitive. Which then begs the obvious question, what’s the harm in the it’s not X it’s Y framework if you’re just going to use AI irrespective.

Ruben Hassid's avatar

the pattern is bad writing - AI defaults to it because contrast feels like reasoning when it is not

using AI badly and using AI well produce different outputs, and the difference is just one anti-ai file

ashivaka's avatar

exactly, you're not trying to fool anyone. If what you're writing about is your original thoughts and you're just using AI to help structure them you don't have to worry about this.

Fabrice Talbot's avatar

Merci Ruben! This is gold :-)

Ruben Hassid's avatar

youre most welcome - did you download the file yet?

Fabrice Talbot's avatar

Yes and I ran a first test on a draft copy. It detected a few issues but missed a few of the parallelism you built in. It also did not catch all the dash. I called it by uploading the http://copy.md file with my content and simply typed audit.

Any advice on getting a more thorough audit?

Ruben Hassid's avatar

did you use this prompt, “Audit it against the anti-ai-writing-style md.”?

Fabrice Talbot's avatar

Just “audit” and I setup the “claude cowork” folder. I did run through. Let me do more passes over the weekend. I trust it works for you so it may be an issue on my side

Ruben Hassid's avatar

hope it works out!! :)

Fabrice Talbot's avatar

Salut Ruben - took a while but I finally validated you anti-ai-writing-style works like a charm. Love it!

ari's avatar

Couldn't have dropped this at a better time - currently testing my skills and was getting so sick of the "It's not this. It's that" every paragraph. Thank you!

Ruben Hassid's avatar

came at the right time then!! let me know how it goes :)

Josh Mclean's avatar

Can it be “Z” 🤪

Ruben Hassid's avatar

well there is the rule of three aha :)

do you notice it a lot?

FORMATION XYZ's avatar

We implemented a similar copy-tone skill. It works great. That and the simple follow up prompt of "do a critical review of the text and address any issues while still keeping the copy-tone in mind as well" does wonders. I still edit a lot manually. But the quality of generated drafts goes up a bit with this.

It would be nice if LLM producers would fine tune their models to stop their models from defaulting to using a lot of hollow/empty language. The Californian origin of many of these models shows in the use of language. Including needing a whole lot of words to say not much at all.

Ruben Hassid's avatar

they are all verbose, hedged, and low-density with lots of disclaimers and filler phrases - its so hard to read aha

until then, my anti-ai .md will have to suffice

Jessica Anslow's avatar

I read this the same day I built a voice file for an AI I work with daily. For me, the cognitive load point is the one that matters most and the one that often gets skipped. Banning the pattern because it sounds like AI is useful. Knowing why it doesn't work on the reader, that your brain processes the rejected concept first and it sticks - that's the more interesting argument.

The pattern is usually a symptom of not having a clear point of view before you write. If you know what you think, you just say it. The reframe is what happens when the AI is generating the shape of an argument rather than an argument.

Ruben Hassid's avatar

you got it - AI has no point of view so it reaches for the shape of one

schatzipage's avatar

I copied the code and entered it into the instructions panel on my free version of claude and pleased to say it worked too!

Ruben Hassid's avatar

thats nice to know - nothing slipped through?

schatzipage's avatar

Doesnt seem that way—I was doing the “interview” last night and it got sassy with me when I used an em dash 😂

Grant W. Petty's avatar

I'm surprised the Anti-AI-writing-style doesn't say anything about the adverb 'quietly'. It seems to be as glaringly ubiquitous in AI writing as negative parallelism!. "X is quietly reshaping how we do Y", "the new model quietly overturned every assumption about what AI could do", etc.

Ruben Hassid's avatar

interesting - i will check on it and maybe add it to my file :)

ManiK Fox's avatar

I was wondering what to call that "it's not x it's y." now I know, thank you. It is so annoying.

Ruben Hassid's avatar

its frustrating because its everywhere