Who writes this glossary

Lexigraph is edited by Anthony Capirchio and published by The Operating Shift. The texts are written with the help of language models, under a grid and writing rules settled before them. What gets published still commits a person: that is the whole point of this page.

The author

And in what capacity

Anthony Capirchio

Designs the grid of the four floors, settles the list of terms, fixes the writing rules and publishes this glossary.

How these texts are written

The AI included

An AI glossary that hid the fact that AI helped write it would disqualify itself. So here is the exact split, with no dressing up.

What a person decides

The thesis of the four floors and where each term belongs. The list of terms kept, and those left out. The writing rules: the one-sentence answer first, the compulsory anti-definition, product names banned from the body of the text. Whether a notion deserves a page, or not. These decisions are written before the texts, and they are what makes this glossary.

What a model produces

The writing, in large part: the sentences of the entries, the analogies, the comparisons on the often-confused pages, the checking questions. A model works here under constraint, in a form settled in advance, on a list of terms it did not choose.

Who answers for the mistakes

Anthony Capirchio, and no one else. The tool used to write a false sentence does not excuse the false sentence. That is why this site carries a name and dated versions, rather than an anonymous footer.

How an entry is made

The method, in plain words

The glossary rests on one thesis: almost every confusion about AI comes from a mixing of levels. Each term is therefore placed on one of the four floors: the model, the harness, the agent, the organisation. This grid is not a presentation of the content, it is the content.

What every entry carries

One sentence: The complete answer in one sentence, always the first block of text
An anti-definition: “What it is not”: the most frequent confusion, taken apart
A technical layer: The distinctions and the traps, for a reader who is building
Dated examples: Everything that goes out of date is confined there, and never in the body of the text
Questions: Answered on the spot, with wrong answers drawn from the graph of relations
A number and a version: The integer never changes, the version gives the month of writing

What the corpus holds today

Entries: 88 terms, spread across the four floors
Often confused: 19 pairs genuinely confused with one another, compared point by point
Infographics: 6 infographics, each with its printable poster
Paths: 4 guided sequences
Under the hood: 12 entries show the literal shape of the object they define

What this glossary is not

The gesture applies to it too

It is not a news feed. No product name, no performance figure, no model version lives in the body of an entry. Whatever goes out of date is confined to the dated examples, so that a text read two years from now is still true.

It is not a tool comparison. The workstations described are types, never brands.

It is not a tutorial. You do not learn how to do things here, you learn how to place them. Even the “under the hood” layer shows the shape of an object, never the recipe for using it.

It is not a translated word list. French is the language these texts are written in, not the language they arrive in: when an English term has no honest French equivalent, the entry says so rather than inventing one.

Reuse

CC BY 4.0

Everything is published under a CC BY 4.0 licence: you may reuse, translate, print and adapt this content, including commercially, provided you cite the source. The citation formats ready to copy are on the “Cite” page, and every infographic gives its embed code.