Mostly setbacks, July 2024

July has been dominated by technical setbacks and incremental progress.

I've focused on developing the wreck incident generator, a system capable of producing plausible maritime casualty statements by combining ship names, cargos, locations, and incident types. Much of this is trial-and-error scripting from extended sessions working with ChatGPT as a coding tool. I'm learning more python by osmosis, but the whole thing is insanely painful. I scarcely know enough of that I'm doing to understand how to make it better, much of the time trying to get ChatGPT to code is like being trapped in a room with an amnesiac. Things randomly get re-written, mistakes repeated over and over. Outputs swing between convincing and absurd but mostly just plain fucking annoying. I feel like the 'easy' thing I picked off, the Lloyd's generator, which I'm hell bent on making, is like some kind of ground zero project,

A recurring challenge was plausibility. Pure randomness quickly undermined credibility. The system required constraint historical tone, vocabulary limits, geographic coherence yet too much control removed generative unpredictability and it just wants to talk about the Titanic. I've a mental image of the lumps or knots in it's model where everything thats not the titanic is just tiny.

I've also a growing sense of distance from the original artistic goal. Weeks producing scripts rather than artworks. It feels stalled. I'm also realising that the biggest immediate problem is the gap between my computer and the world. I'm going round and round in circles trying to get stuff onto the web or get stuff onto a printed stream. None of it's working. For all the creeping ability to work with scripts the dev ops aspect is entirely maddening.