No. 122 — 25 454 Links ⫶ He Did It in French ⫶ Kompyuuta vs. Tölva

Not to destroy wonder

My name is Linda. I write a bi-weekly newsletter about computer science, childhood, and culture.

Last week I attended one of my very first French dinner parties. In Finland, a dinner party is a series of one-on-one conversations. In France, it's a different social technology entirely: the salon is a four-hundred-year-old invention and operates under completely different rules.

The hosts conducted the room. They knew when to throw a question to the quiet end of the table, when to introduce a little friction, when to reframe what someone had just said so it opened to the whole room. It was magic to experience (and also an amazing three hour listening exercise for my French).

At some point, someone said (perhaps joking, perhaps not) that no real philosophy is done in English. Someone else yelled: “What about David Hume?” The first person waved his hand. “He did it in French.”

He didn’t, of course. But the Parisian table had claimed him anyway, which is its own kind of argument.

1.

I’ve been thinking about that exchange all week because I think it’s actually true, at this particular moment in computing.

The claim (no philosophy in English) is a real position in French intellectual life, and one I used to find more than a little self-satisfied. The argument runs roughly like this: what English-speaking academia calls philosophy is actually something else. Logic, probability, philosophy of language, philosophy of science.

But that’s not la philosophie in the French sense. I wouldn't have cared, except that this year I keep bumping into the same problem from the other side.

The architecture of large language models was built entirely on the analytic tradition's home turf. Probability, information theory, Wittgenstein's language games. Shannon, Turing, Russell, Wittgenstein: they all treat language as structure.

And yet what has emerged with large language models is something the analytic toolkit can’t quite explain. When I ask “does it understand?” or “is there someone in there?” I walk straight into the questions the French were asking seventy years ago. The continental tradition treats language in a completely different way. Saussure, Baudrillard, Derrida, Foucault - they all insisted that language doesn't describe the world, but constructs it. I don’t know enough about the French tradition to say where it takes us, but it makes me want to read more Foucault and fewer benchmark papers right now.

2.

This editions illustrations come once more from Dana Stewart’s computer science class at Catherine Cook School in Chicago (here is work they did on machine learning and training data). I got to meet the class over Zoom the other week and they were the best workday ending I could have asked for. Dana’s questions for the children: what does the Internet look like? What networks do you belong to? And how will Internet look like in 100 yeras?

And for a small activity on what the internet looks like, check this out.

3.

Riley Walz built a tool called In Every Language that shows the word for any concept across different Wikipedia languages, alongside each edition's chosen image. Here is the word "computer." The translated words alone are fascinating. Some languages borrowed the English ("komputer," "kompyuter," all the way to Jamaican Creole's "Kompyuuta"). Others invented their own: Finnish gave us tietokone (knowledge machine), Icelandic tölva (number prophetess), Chinese 電腦 (electric brain), Turkish bilgisayar (information counter). Each language had to decide, sometime around the mid-1950s: is this thing a calculator, an organiser, a brain, or an oracle?

5.

Made me smile: a D&D alignment chart for the verb you use when you're making something. Creating, crafting, constructing, concocting.. It reminded me of how Claude Code apparently shows fifty-seven different, playful and verbose status words while it works on something (Ruminating, Schlepping, Moseying, Honking). Cowork, the mode I use more often, shows plainer ones: "Searching newsletter archive," "Extracting quotes." Every one is a retrieval verb.

Claude Code is a tölva. Cowork is a tietokone.

6.

I spent a weekend building a personal Wikipedia from eighteen years of bookmarks. Like seemingly everyone else, I read Andrej Karpathy’s note on LLM knowledge bases. This is something my collector brain absolutely loved (and I’ve written about the importance of small archives before). Karpathy’s insight is that an LLM knowledge base isn’t only a search engine, but a way of having a conversation with your own past.

The raw material: my bonkers tendency to save links, notes, and remarks from 2006 to 2024 across del.icio.us, Kippt, Pinboard, and Are.na. 25 454 links, to be exact. The ambition is to turn this scattered archive into something I could actually think with and connect the dots I’ve been collecting since I was twenty-three (!!). Gosh, it's a fun time to be thinking.

5.

“Stevin was so pleased with his construction that he incorporated it into a vignette, inscribing above it Wonder, en is gheen wonder that is to say: “Wonderful, but not incomprehensible.” This is the task of natural science: to show that the wonderful is not incomprehensible, to show how it can be comprehended—but not to destroy wonder. For when we have explained the wonderful, unmasked the hidden pattern, a new wonder arises at how complexity was woven out of simplicity. The aesthetics of natural science and mathematics is at one with the aesthetics of music and painting—both inhere in the discovery of a partially concealed pattern.”

- The Sciences of the Artificial (1996) by Herbert A. Simon.

Not to destroy wonder! Could be from a Reggio Emilia lecture, is from one of the most ageless computer science thinkers, my very favourite Herb Simon.

6.

Tivadar Danka is at it again. This is the Marauder’s Map of my dreams: a machine learning knowledge graph with 2081 concepts and 5149 connections between them. You swirl around the graph, find connections, paths, and rabbit holes you didn’t know existed. Click on “ergodicity” and it shows you the 41 concepts you need to understand first. Makes my little computer science map seem ancient!

7.

There is a new Mary Oliver documentary coming out this summer. Here's a list of upcoming screenings. I still love this episode of On Being where Krista Tippett mentions that her child, at twelve, was assigned to memorise "The Summer Day." I plan to do the same with mine. One exchange from that conversation:

Tippett: I was going to ask you if you thought you could have been a poet in an age when you probably would have grown up writing on computers.

Oliver: Oh, now? Oh, I very much advise writers not to use a computer.

Tippett: But it seems to me that more than the computer being the problem, the sitting at a desk would be a problem.