Reflections on Working with LLMs

I’m noticing a shift as I work with LLMs. I used to manipulate my code as if sculpting angels in the architecture. Each program was a cathedral housing ornate beauty. The code I generate with LLMs is an airport parking structure. Pure utility with the necessary materials.

Both structures have their purpose. Sacred spaces for quiet reflection serve my needs. Protecting my wheel vehicle while I travel in a wing vehicle also meets a need.

I’m witnessing a realtime rebalancing of my values. I’m following curiosity faster. I’m delaying craftsmanship until there’s proof of utility. This is a positive change for me. I tend to get stuck in prematurely predicting what a solution will need rather than attempting to produce a system I want. Yes, code is just another avenue for humans to express and distinguish their wants and needs.

I want infrastructure for making human connection. LLMs for building pipelines feels right. LLMs clogging those pipelines with slop isn’t interesting. And yet, “the medium is the message.” The medium is produced by the statistical average of the next token that should appear. Perhaps it’s modernized as “the mean is the message”—the computed average of logarithmically accelerating messages received is how we interpret the world, and decide what to do next.

Man’s search for meaning (pun intended) should not decline to the mean because of this new mode of operating. Satisfying the search for meaning requires choosing, not averaging. I’m using the means of production to build infrastructure, while keeping the search for meaning for myself. The parking structure exists to get me to the cathedral faster. Maybe the real cathedral was the curiosity I followed along the way.

I’m building this personal website with the help of LLMs. I wrote these words with my own hand using Bear for iOS and Bear for macOS. You long press the dash to insert an em-dash on iOS. You chord option-shift-dash to type an em-dash on macOS. Just because the computer predicts em-dashes too often doesn’t mean it’s not the character I want.

Acceleration of discovery, not delegation of meaning, is the key.