How I Write
🪨 How I Write Articles (According to Me, and Two Machines)
People sometimes ask how I write the essays on this blog.
The short answer is: with a notebook, a stubborn idea, and a small panel of
artificial judges who are not afraid to tell me my idea is unoriginal. The long
answer is below.
🧠 Step 1: Pretend I Have an Original Idea
I start by convincing myself I’ve had a brilliant, never‑before‑seen
insight. To maintain the illusion, I write a short description of the idea,
followed by a bullet list of related topics, issues, and things I will probably
forget to include later. I also write down counterarguments, because nothing
says “healthy creative process” like arguing with yourself before you’ve
written a single sentence.
🤖 Step 2: Summon the First LLM
Once I’ve built enough confidence to risk disappointment, I
tell an LLM that I’m “thinking of writing a magazine article” about this topic.
This is code for: please don’t embarrass me in front of myself.
I then ask the machine the most important question in my
workflow:
“Is this an original idea?”
If the LLM says “no,” I immediately abandon the topic, as
any reasonable adult would. If it says “maybe,” I take that as a resounding
yes. I also ask it for more counterarguments and possible titles, because
nothing sharpens a thesis like a robot telling you why you’re wrong.
📝 Step 3: Produce a Draft (Human Participation Optional)
After enough back‑and‑forth, a draft appears. I like to
think I wrote it, but the LLM and I both know it was a joint custody situation.
📄 Step 4: The Second Opinion
I then export the draft to a Word document, convert it to a
PDF—because nothing says “professional writer” like unnecessary file
formats—and upload it to a different LLM. This second model has no idea
what the first one said, which makes it perfect for unbiased criticism.
I ask it to review the argument, critique the writing, and
once again answer the existential question:
“Is this an original idea?”
If both LLMs agree it’s original, I celebrate. If they
disagree, I assume they’re fighting and publish the article anyway. So far, they have never disagreed.
🎨 How I Create the Images for Each Article
Every article on this blog comes with an illustration, and those illustrations follow a very strict artistic tradition: cave people interacting with modern concepts in completely absurd ways. If I can’t think of a prehistoric gag that fits the topic, I take it as a sign that the article may not actually belong on this blog. This is my version of peer review.
🪨 Step 1: Come Up With a Cave‑Person Scenario
Once I have the article’s theme, I try to imagine how a cave person would misunderstand it, misuse it, or accidentally invent it. If nothing comes to mind, I question the entire premise of the article and briefly consider switching careers.
🤖 Step 2: Ask an LLM to Help Write the Prompt
When I do have an idea, I ask an LLM to help me craft the image prompt. I make it very clear:
“Help me write the prompt. Do not create the image.”
Then I describe the constraints:
- Flat white background
- Black ink, clean lines
- Minimalist, New‑Yorker‑ish style
- Cave people doing something ridiculous with a modern concept
The LLM doesn’t generate the image—it just helps me refine the specification. Sometimes it suggests adding a detail I hadn’t considered. Sometimes it suggests removing a detail I definitely should have considered. After a few iterations, the prompt becomes oddly precise, like a police sketch description of a joke.
📝 Step 3: Ask for the Final Prompt
Once the specification feels right, I ask the LLM:
“Write the final prompt to create the image.”
This produces a clean, ready‑to‑use prompt that I can feed into whichever image model I’m using that day.
📁 Step 4: Save the Prompt for Future Mischief
I save the prompt alongside the article. This is partly for archival purposes and partly because I enjoy trying the same prompt on different models to see how each one interprets “cave person discovering blockchain.”