Why I'm Slowing Down

(and why that’s a good thing)

I’ve been living on social media for the past ten years, and several times a day, I pick up my phone or open a new tab in my browser and check the fresh content my timeline throws at me.

Consuming content never felt as terrible as it feels now. Given the various “metas”¹ of the past couple of years, that’s quite a statement: we’ve seen

• misogynistic red-pill shorts and Andrew Tate clips flooding every platform,

• superstars shilling shitcoins for money on every available outlet,

• or watched the chaos unfold during the first wave of COVID, with millions launching content creator careers from their quarantines.

(If I ever launch a punk band, I’ll name it Superstars Shilling Shitcoins.)

Today, even though it doesn’t feel as dramatic, I can’t help but constantly feel that something is off. Sure, we all knew that the President of the United States minting his own crypto was absurd; it was a Black Mirror episode happening in real life.

Right now, it’s more like Pleasantville, as my wife pointed out. She has been obsessing quite a lot lately over the model collapse phenomenon.

If you don’t remember Pleasantville, let me explain. I wouldn’t have remembered it either if it was not the very first movie I watched on HBO in the early 2000s. Growing up in a post-socialist country like Hungary, HBO was like a gateway to the Western world. So, this movie became a foundational memory for me.

In Pleasantville, everything is pleasant. (Surprise.) People are nice. They politely greet each other when they meet in the mornings, noons, evenings. The handsome husbands go home straight after work, just in time for dinner perfectly served by their gorgeous wives. (In the always spotless kitchen.) The high school basketball players never miss shots, fires are impossible to start, and the firemen’s only job is to rescue cats stuck in trees. Dates always end well, with lovers holding hands, staring at the stars in the night sky. Pleasantville is the fictional version of LinkedIn. Oh, and everything is black and white. No pain, no pleasure, no lows, no highs. It’s all predictable. All average.

Why are we talking about this? Because reading AI-generated content feels like spending time in Pleasantville.

Sentences are well-formed. The grammar is spotless. The voice is polite. The content is designed to avoid challenging your beliefs, because that would perhaps make you stop reading. No swearing. No sarcasm. No controversial opinions.

Really, it’s the nature of AI-generated content. Large language models generate responses token by token, selecting the one that has the highest probability given the previous output tokens. In other words, at their hearts, LLMs are statistical bullshit generators.

They are trained to say what you want to hear. Not what you need to hear.

(Yeah, I know, there’s reasoning, tool calling, memory, and other neat tricks. Still, fundamentally, generative models are “stochastic parrots”.)

For functional purposes like building makeshift apps, they are unbeatable, though. I don’t want my Python scripts to challenge my beliefs; I want them to work. I’m using them to automate tasks and build things. If I want a short animation illustrating gradient descent, I don’t care how neat the Manim script is. I only care if it does what I ask it to do, and that’s it.

LLMs are also perfect for collecting and processing information. Whenever I want to quickly learn about a technical subject, such as video game architectures or epidemiological models, nothing beats chatting up Claude and going from zero to one in a couple of minutes. Hell, there are Substacks with LLM-written posts that provide a ton of valuable information, ranging from science to parenting through engineering and education. I’m even using LLMs to fix the grammar mistakes in this very post you are reading.

But that’s not what I want to read when I subscribe to you, and I’m guessing that this is not what you, who are reading The Palindrome, want from me either.

I’m reading Substack to find something that I cannot find in a discussion with a chatbot. I want you, the author, to make me uncomfortable. I want to glimpse into your beautiful mind, your unhinged thoughts, your limitless creativity. To question my beliefs, to see old things from new perspectives, to expand my horizons.

To update my weights.

Lately, it’s hard to do. I open my timeline(s), and it’s full of bland, uninspired, fortune-cookie bullshit. Every note and post is formulaic. Everyone is desperate for engagement, but no one wants to engage. I stopped reading Notes. I stopped writing them. I reached a point where I’m reluctant to read books released after 2022.

As a content creator, I’m facing a choice now. One: I automate the writing and publishing process with AI to keep up the pace. I’ll feed it all I ever wrote (a couple thousand pages already), make it learn my voice (Eastern European mathematician, loves death metal² and gardening), give it a couple of topic ideas (machine learning, mathematics, growing up in a post-socialist country right after the collapse of the Soviet Union), and have it churn out post after post, feeding the content machine.

Two: I make something that’s better than 99% of what’s out there. Learn to tell great stories. Bring my words to life via beautiful animations. Take the time to execute grand ideas instead of publishing mediocre ones twice per week. To improve my skills, and use AI to augment them instead of replacing me. To be the signal, not the noise.

To reach for quality, instead of rushing for quantity.

If you have been following me for a while, you already know which path I’m taking. This year, I put growth on hold and spent all my work hours increasing the value of my content from every aspect.

I moved from static illustrations to rich animations.

I started to build interactive visualizations that you can explore along with the posts.

I learned the foundations of audio engineering, invested in a high-quality recording setup, and took a couple of voice coaching classes to improve my voiceovers.

When I’m not creating, I’m reading about writing, storytelling, and animation.

I’m becoming an artist. Not a content creator, not a technical writer, not a Substack author.

For me, nothing can replace the joy of creation. I’m writing these lines, rewriting them, picking them apart and putting them back together, second-third-hundredth-guessing every sentence, changing the pace, playing around with its rhythm, converging toward a chunk of text that has me within the lines.

For me, writing is thinking, a process where I zero in on the truth. (See my Coding on Paper essay, where I talk about this.)

I won’t automate my existence away. In the future, you’ll get more of my thoughts, instead of just more words, shaped by an algorithm.

---

Thanks for reading! In a regular newsletter issue, this would be the place for a Call To Action, a message to convince you to subscribe, engage, etc. Instead of shoving you down a marketing funnel, I’m encouraging you to do some thoughtful introspection.

If you are a content creator: do you use AI responsibly, or is your channel just the Substack equivalent of dropshipping?

If you are a reader: if your body is what you eat, then your mind is what you read. What do you read?

1. meta is short for “most effective tactic available”, a term familiar from competitive games. Not picking the meta strategy automatically puts you at a disadvantage, usually narrowing players into particularly annoying playstyles.

2. My OCD told me that “one footnote is not a footnote”, so I added one at random. Sorry for that.