Using an LLM to avoid looking like I’m using an LLM

I mentioned that I want to avoid the “Claude aesthetic” in this blog’s design. It’s not that I have anything against it, really, or against AI-built websites as a concept.

My hope, though, is that this AI frenzy provokes a renewed appreciation for handcrafted things, and I want this site to be a vote for that in some small way. And so, with a heaping megadose of irony, I used an LLM to help me make this blog look less LLM-generated.

The choices, for now

First, I went for the paper-on-desk look. I deployed it a week or two ago. If you noticed … how nice of you!

But it felt a little stuffy, or boring, or something. So I explored some different colors and textures. I tried a dot-grid texture on the background and liked it, but I realized it was almost exactly what Kev currently has, so I moved on. I tried a “wool felt desk” look, but it was too cold, and the texture was a bit too fine, almost unnoticeable. I tried some higher-contrast options, but the background was too dark for light mode.

I landed on a “linen weave” texture with a warm-toned background, vaguely similar in color to what I was already using, which itself was vaguely based on Gruvbox’s light mode but less aggressively yellow.

It doesn’t spark joy quite yet, but it’s good enough for the moment. If you see any bugs or gaffes as I flop around, please email me!

Why an LLM?

I consider myself a creative person in some respects, but not at all when it comes to visual design. That’s why I chose a flat design in the first place: fewer decisions to make, less spinning of wheels. When I try to get fancy, I tend to burn lots of time and get mediocre results.

With an LLM, however, I can go from a general concept, like more humanness or skeuomorphism, to specific ideas in seconds. It implements those ideas so rapidly that I can literally type “next” until I see one I like. And then it can tweak them based on my vague nudges like “warmer background color” or “more subtle texture” until it feels close enough to good. Meanwhile, I can direct all of this activity while maintaining a conversation with my wife. That’s wild.

This doesn’t mean I’m pro-LLM. I still wish I could magic-wand them into oblivion. But since they’re here to stay, I can admit that they’re really good at stuff like this.

Back to the title of this post: does this make me a hypocrite? Maybe, but I don’t think so. If this were a flashier, more design-forward blog, or I were somehow implying that I’m a “design guy” when in fact the LLM did all the creative work, then yes, that’d feel hypocritical. Maybe that’s what bugs about the Claude aesthetic — it’s trying to look hip. But hip is a human thing, so it feels phony.

In contrast, I hope it’s obvious that I’m not going for any design awards here. I just want something that’s easy to read and unique enough to distinguish it from off-the-shelf designs or AI defaults. I think I can use an LLM to facilitate that effort without treading into hypocrisy.

I’d say I’m more like Don Draper, sending the underlings off to generate ideas, and applying my cunning wit to select and refine the best one.

Yes, that’s exactly it. I’ll go tell my wife.