⌘ AI

How I use — and don't use — AI/LLM tools

Preface, or my distilled thoughts on AI

I don't think we were designed to stomach the effects of this technology, at least not at the scale and rate we're seeing. I was torn when the ordeal took off in late 2022, and I'm still torn today. With that said, here is where I stand on the matter (for now).

I recognise that AI is causing significant harm. It's changing a common landscape that took us centuries to build. (It may not have been pristine, but it's deeply rooted in us as a species.) AI is shifting our thought patterns and behaviours. It's leaving students confused and professionals jobless. It's stealing from millennia of art and innovation. It's making us efficiently lazy. It's causing decision-making atrophy. And so much more.

But I'm not a Luddite; I recognise that not all AI is bad, and that some of it can help. And since it's here to stay, I believe a complete abstention is a little extreme and will likely set you back to some degree. However — big caveat — I'm very aware that using AI can fundamentally affect critical thinking, which is perhaps the biggest attribute I rely on for my work. That's why, despite being open to its use, I tread carefully and limit my use.

These days, AI companies are investing heavily in making their products cute, fuzzy, and warm. They're trying to bridge the gap between "mysterious powerful data centres" and the average, terrified Joe. I disregard these efforts; I don't anthropomorphise an LLM. I refer to one as "it". That's not to say I'm unkind to it. I try to speak to chatbots as I would to a human — not because I feel for them, but because I want to practice kindness.

Instead of a friend or a writing partner, I view this technology as a (highly capable) tool or service that can get some of the trivial/mechanical/tedious/admin tasks out of the way faster, so I can focus on what truly matters: creation.

I'll sum up my AI usage with an analogy. I can clean my house, but it doesn't directly contribute to my daily word count; it deters me. Yet it must be done so I can work comfortably and be a good host. That's why I have no problem hiring a cleaner to take care of it. Now, if the cleaner were to offer to "help" me come up with some good writing, I'd draw a line — with permanent ink.

How I don't use AI

If there's anything I want you to take away from this page, it's this:

AI never writes anything for me.

I value my voice. It's me with all my virtues, foibles, and mysteries. I've spent years finding and alchemising my voice, and it's taken me another good odyssey to find the courage to use it. I'm still improving. I'll always strive to improve. That's why I would never give up my voice for convenience or more output. Compelling work takes time. I don't mind that. It's an organic law. Let me be a snail!

Even if one day AI can be trained to generate 99,99% in "my voice", even if it "transcends" 100%, my interest would be zero. Writing isn't a burden; it's the whole point. Writing gets me out of bed in the morning. It's my art, goddammit! Having an LLM generate something from a trillion-fold average and calling it "my work" would completely violate my values as an artist and leave me without a purpose. No, my imposter syndrome is already high, thank you very much.

I'll always be the one stringing the words together. It'll always be me painstakingly ideating, structuring, and weaving character and plot into a narrative. Writing is my jam.

This statement extends to any other outlet where I write as myself — namely emails, comments, and instant messages.

How I use AI

As far as I'm aware, I have active accounts with two AI/LLM tools. I'm on a free plan with both and use them for the following purposes:

Research

When I'm writing about something I'm not familiar with, I might ask an LLM for an overview or answers to concrete questions. Depending on the answer, I might browse the sources or ask for others. I don't recycle any of the text; I absorb it like I would by reading a blog, a book, or a newspaper.

Opinion

When I'm hesitating, or weighing two or more options (in a non-creative, generally low-stakes context), I might ask an LLM for an opinion. Writing-wise, this can be grammar, formatting, planning, and cultural sensitivity, among others. I'll often dig deeper by raising a concern, or even challenging it. It's a dialogue.

If the matter is serious, I'll consult my family, friends, and professionals. Maybe — in the final stage, just for fun — I'll ask an LLM for a fourth opinion.

If the matter is creative, I don't use an LLM; I use my noggin. When I hit a wall, I don't resort to an LLM; I take a walk and repeat the process.

Proofreading

This is my "anti-stupid mistake" safeguard for my non-fiction (the things on my website). I primarily write in English. Despite it being my strongest language, it's not my native tongue, so I welcome prods to grammatical mistakes and misspellings (I still can't type "miscellaneous").

Another thing that gets flagged during this process is typos. As pedantic as I can be and as much as I review my words, I surrendered to the typo war long ago.

My fiction, however, is sacred. It only gets the barebones treatment: the Spelling and Grammar feature on macOS. I don't proofread it using AI.

Code

I'm not a programmer. I've never studied code. I can barely wrap my head around the difference between HTML and CSS. I know what UI and UX are, but don't ask me about Python, stacks, or frameworks — no clue.

I'll resort to AI for the occasional "code injection", which essentially overrides a specific piece of default code on my website (my CMS is Ghost). These are mostly minor aesthetic tweaks — rounding the corners of a block, changing the background colour of a page, and rare specifics like "make it so that anything on this page that's within brackets has a lower transparency".

Developers, I hear you laughing, and I laugh with you.

That's it, really. I'll continually update this page as my AI usage ebbs and flows.

Oh, and if you're here looking for inspiration to create your own AI usage policy, don't stop here and consider the many other examples.