Some thoughts on AI in general and this site specifically

AI and generated content are all the rage (a sometimes just TEH RAGE) these days. I’m generally not a fan: I know enough about how these systems work that I’m not sure there’s much real artistic generation going on, just some statistical output based on some input. Sometimes it looks cool. Usually not. Sometimes it is useful.

Some Motivations

When I set out to build this website, my intent was to build a representation of myself and my work on the Internet. This is partly because I feel that I need some professional presence, and partly because I hope to share some of my photography, and partly beause I want to get into the habit of documentating and even publishing some of the projects that I consider interesting in the hopes that they’re interesting to others.

I’m not a great web developer. Most of my web skills and graphic layout abilities were defined back in the earlier days of the Internet. As in, 1994, when it was normal to send an email to Switzerland saying “I built a website, please add me to the web.”

While I generally know what is going on under the hood here with HTML and CSS, I really have no interest in actually doing it or learning how to do it. I could pay someone to build a website for me. I could also just use a template from Squarespace or whatever. Right now I have an AI system building out “code” for a system called Astro to use to generate these pages. Personally, I don’t really consider this “generative art” any more than the transpiler that converts the TypeScript into JavaScript or the virtual machine that executes the JavaScript to tell the browser what to render.

The Rules Around Here

A future essay, when I get around to writing it, will probably discuss my current workflows, and how I’m building things, but for now I’ll just include the rules I added to the AGENTS.md file for this site.

The intent is that this website will represent my personal work, in the form of my
(human readable) writing and pictures. While AI or machine learning or LLM systems
can be used to generate code and do page layout. If there is any "presented"
content that is generated by a computing system, it should be reconigzable as such,
so that it can either be identified by end users or by replaced by human work
during proofreading. This means that text content should be the traditional "lorum
ipsem" or other similarly identifyable material, and that images should be
obviously generated: scribbly cartoons or outlined in red or overlayed with warning
stripes or such.

So far this is working pretty well: the agent has been keeping track of the assets in the site with a checklist so I can make sure to back and find or generate legitimate assets before publishing. The “Lorem Ipsem” treatment makes it easy for me to identify text.

I’m still using the same ML-based spellcheckers that everyone uses, and I’m willing to ask the robot to estimate how long it will take to read this. But the content on this site, even down to the icons, should be mine, even if I’m using tools to do the generation of the system that does the formatting.