Marginalium is an uncommon singular form of marginalia.1 You might have come across medieval marginalia online: drawings in the margins of manuscripts, some of which are pretty bizarre. There is the penis tree, for example. There are also creatures getting shot in the butt with arrows.3 I have my reservations about BuzzFeed, but they do have a collection of these, and apparently that's enough for me to cite them in my first blog post.

The written notes are interesting for a different reason. Book historian Erik Kwakkel describes how medieval scribes and readers left comments that help us understand what they thought of the text. Sometimes they corrected it. Sometimes they complained about a bad translation.4 You have the main text, and then you have someone in the margin who has a few things to say about it.

That's the part I like. It's tempting to call them medieval shower thoughts, although that would be taking some liberties with the history. I'm borrowing the idea, not claiming to know what someone was thinking when they drew the tree.

Each post here is one margin note about AI. Something I've been thinking about while using it, building with it, or trying to make sense of whatever is going on in the space.

How I ended up here

My connection to AI and machine learning is mostly being an early adopter who kept finding things to do with it. I don't have a PhD in applied mathematics or computer science. I do have a Python script from 2023 and a lot of time spent using these tools, which is a somewhat different set of qualifications.

The script was Fresh RSS News summarizer, the early version of what eventually became REED. I wanted it to take news from an RSS feed and summarize it in a consistent format. It used preset instructions and could call either OpenAI or a local model through Ollama, rather than having me do everything in the ChatGPT web app.

Looking back, I'd call it a very basic harness: some code around a language model that gave it a particular job. There was also a RAM and CPU monitor intended to shut down the app and inference if resource usage got too high. Running the model on my own computer came with some practical considerations beyond whether the summary was any good.

I wrote it by hand while watching a bunch of YouTube tutorials. At the time, I didn't find LLM-generated code reliable enough to depend on, so I was learning how to build with AI from humans explaining things on the internet.

These days, I use various AI agents to help organize my ADHD life and work through project ideas. There's MachineRead, and then there's a library's worth of private repositories on my GitHub. I realize linking you to things you can't see is not particularly helpful, but that is where a lot of the experimenting happens.

What I'll be writing about

Anyways, I spend enough time with this stuff to have a lot of thoughts about it. My understanding of how large language models work is still limited, but probably a little more informed than average from actually using them. This blog is where I want to work through those thoughts without pretending I'm writing a research paper.

I'll keep the jargon to a minimum. If I need a technical term, I'll explain what I mean by it. I'm trying to have a conversation about this stuff, not give either of us homework.

Honestly, watching YouTube videos about AI can be painful sometimes. Too many sound like someone had ChatGPT or Claude write the entire script. I came to hear what a person thinks about AI, and somehow I'm listening to AI explain AI through a person. With all the same buzz phrases.

So I'll be writing these posts myself, with help on the editing and fact-checking. There will probably be tangents. The first drafts will definitely contain paragraphs that should have been several paragraphs. But I'd rather start with something I actually want to say and clean it up than have a model decide what my thoughts should sound like.

If that sounds like your kind of AI blog, stick around. I have a lot of notes.

Sources

• [1] Wiktionary: marginalium

• [3] Baby Praying Mantis: 20 Bizarre Examples Of Medieval Marginalia

• [4] Erik Kwakkel: Getting Personal in the Margin

• [5] Ashish Vaswani et al.: Attention Is All You Need (2017) — Source of the research-paper text reinterpreted in the cover image.