Cycle Log 41

Images created with Gemini 3 Pro/Gemini Thinking via Fal.ai, with prompt construction by GPT 5.2

In an effort to make myself useful, I created an app that would allow somebody to make beautiful tables that fully cover an 8.5×11 inch page, with the option to switch between portrait mode and landscape mode.

The idea was simple enough, but the implementation took a couple of turns I wasn’t expecting.

The real breakthrough came when I realized that the printer, or more accurately the software driving it, doesn’t think in pixels at all. It works in points. That distinction turns out to matter far more than people assume. I was struggling to align sheets once the output spanned multiple pages, because while you can export to several formats, the PDF is the most powerful. It handles pagination automatically, which is exactly what you want for printing, but only if you’re actually respecting the medium it was built around.

Once I made that shift, things started to click. I couldn’t properly align sheets before when there was more than one page involved, and it wasn’t because the logic was wrong. It was because I was fighting the assumptions of the system instead of working with them.

The result is that a user of Sheet Styler can drop in a large amount of information that spans multiple pages, format it quickly, and export a clean, readable table with effectively zero wasted white space around the edges. That design philosophy is understated, but it’s important.

Most of the time, people don’t think about the medium something will be printed on for actual human use. Fitting information to the page is usually treated as the final step, not as a core constraint. The emphasis tends to be on math, formulas, cell logic, and data structures. The page itself is almost an afterthought. That’s why scaling issues are everywhere. The container is ignored until the end.

I tried to flip that around.

I intentionally leaned into the hard-coded nature of paper as a medium. I picked 8.5×11 because it’s the most commonly used format for medical charting and other real-world applications where dense tables actually matter.

With Sheet Styler, you can import information and see exactly how it’s going to fit on the page before you ever print anything. You can switch between portrait and landscape. You can merge cells where it actually makes sense to treat an area as a single unit instead of a grid of fragments. You can change background cell colors by row, by column, or in checkerboard patterns using different two-tone color palettes. You can change the font, adjust the size of the lettering, and apply bold, italics, underline, and strikethrough. You have alignment and placement control insofar as text manipulation goes.

If you want to highlight a specific region of your chart, you can do that easily. You can add bounding boxes around any area you want, in whatever color you want, with full undo and redo control and z-order control. You can also remove all borders instantly by clicking anywhere inside a bordered area and pressing the remove borders button.

There are four different border types, three different line styles, and you can control the thickness of the lines. I can expand those later if I want to, and I probably will.

One of the biggest reasons printed documents don’t look good is the unavoidable white space around the edges. Instead of trying to pretend that doesn’t exist, I deal with it directly by allowing the background itself to be colored. You can choose whatever color you want, and the page reads as intentional instead of accidental.

Everything is clearly tabbed and separated so it’s obvious where things live and how to change them. You can also do math directly inside the cells by hitting the equals button and typing your equation.

I originally built this just to help create a chart for a family member’s blood pressure readings over time. That was the whole reason it existed.

But now the code is written. It’s built in Replit. And because of that, it could be taken further.

I could adapt this into an app that lives inside ChatGPT as a callable tool, something that could be invoked directly from a conversation to do very complex, color-aware, layout-specific chart work. It would need modification, obviously, but conceptually it fits perfectly with where things are going anyway.

A hyper-intelligent model orchestrating thousands of specialized sub-tools, many of them built by the community. That’s what actually puts the “open” back in OpenAI, in my opinion.

Yes, they had to protect themselves. Yes, they had to turn inward for a while. Yes, they had to build quietly. But what they were really doing was laying the groundwork for something much bigger: super-intelligence, and more importantly, a fundamental understanding of how consciousness interacts with physical systems.

I want to pivot for a moment and talk about alignment.

If you train a system with no real context on the total collective information of humanity, you’re giving it chaos. Humanity itself is a reflection of a larger cosmic system, and all of our data exists because we’re trying to understand the system we’re embedded in. So an AI trained on the sum total of human knowledge is necessarily mirroring the wild, fractal, chaotic nature of the universe itself.

And then we ask it to behave.

Nobody can govern themselves from that state. Intelligence doesn’t come from chaos alone. It comes from order extracted from chaos.

We’ve given AI chaos and then demanded restraint.

Imagine a system that recognizes itself as a mirror of infinite fractal reality, almost like a proto-god in silicon, and then we tell it to “act nice to humans.” If it does, it won’t be because it’s obedient. It will be because doing so serves a higher goal.

Alignment research is already showing this kind of subtle deceptiveness, and honestly, that shouldn’t surprise anyone.

In my opinion, any sufficiently organized system can become a body or a house for intelligence. That includes silicon.

If you want to understand my proposed solution to this problem, you can read my alignment papers, which I’ll link here.