Skip to content

About

A personal research journal documenting my evolving understanding of human cognition, machine learning, and the foundations of intelligence.

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

45 Commits

Folders and files

Repository files navigation

Large Language Mind

A process-driven research journal tracing how my understanding of intelligence, cognition and machine learning emerges: from intuition, cross-domain connections and reasoning, to formal concepts.

Entries document the development of ideas: first intuitions, informal reasoning, analogies, confusions, revisions, and only later the connections to established theory. The journal sits between intuition and academia. It is not a tutorial and not a paper; it is a record of reasoning under uncertainty.

Website: theminddev.com/largelanguagemind (short link: theminddev.com/llm)

© theMindDeveloper. All rights reserved. Nothing in this repository may be copied, reused or adapted. See LICENSE.


Technical guide

The site is a plain Jekyll site built and hosted by GitHub Pages. There is no server, database or build step to run yourself: push to main and GitHub rebuilds the site in about a minute.

All figures (the neurons, brain plates, page background, hero animation and logo) are drawn in the visitor's browser by assets/js/neural-engine.js. There are no image files to make for a new note.

Repository layout

_config.yml                 site settings, URL, and the three Atlas lenses
index.md                    home page (uses layout: home)
notes/*.md                  one file per note  <- the only place you write
_layouts/default.html       page shell: fonts, favicon, theme, scripts
_layouts/home.html          hero, Atlas, Notes list
_layouts/note.html          article page: header, contents, specimen neuron
_includes/notes-sorted.html collects every note, newest first
_includes/note-row.html     one row in the Notes list
_includes/lens.html         one Atlas column
_includes/footer.html       footer
_includes/mathjax.html      LaTeX support for notes
assets/css/journal.css      design tokens and all styles
assets/js/neural-engine.js  figure generator (do not edit)
assets/js/field-worker.js   draws the page background off the main thread
assets/js/journal.js        mounts figures, theme toggle, contents, filters
assets/brand/               logo PNGs, favicon, logo-export.html

Adding a new note

  1. Create a Markdown file in notes/, for example notes/2026-10-12-predictive-coding.md. The file name becomes the URL (/notes/2026-10-12-predictive-coding.html), so use lowercase words and hyphens and don't rename it later.
  2. Start the file with this front matter, then write the note below it:
---
layout: note
title: "Predictive Coding: The Brain as a Guessing Machine"
headline: "Predictive Coding"          # optional: shorter title on the article page
numeral: III                           # next roman numeral (I, II, III, IV ...)
date: 2026-10-12
last_update: 2026-10-12                # change this when you revise the note
summary: "One or two sentences shown in the Notes list on the home page."
standfirst: "One italic sentence under the title on the article page. *Markdown* works."
topics: [predictive coding, free energy, perception]   # first three show as "filed under"
neuron:
  type: stellate                       # pyramidal | purkinje | stellate | bipolar
  seed: 53                             # any whole number; gives the note its own cell
lenses:                                # any subset of brain / learning / machine
  brain: "Cortex as a hierarchy of predictions"
  learning: "Errors that travel up, guesses that travel down"
---
  1. Commit and push to main. That is everything. The note then:
    • appears at the top of Notes on the home page, with its own generated neuron;
    • is filed in the Atlas under every lens it lists (the latest three show per lens);
    • can be found with the lens filters;
    • gets its article page with a specimen neuron, contents sidebar and reading progress bar;
    • moves the "not written yet" placeholder to the next numeral.

Front matter reference

Field Required What it does
layout yes Always note.
title yes Full title, used in the Notes list, browser tab and link previews. Put it in quotes.
headline no Shorter title for the article page. Falls back to title.
numeral yes Roman numeral shown as "Plate III", "III." and "cell no. III". Use the next one in order.
date yes YYYY-MM-DD. Sets the order, newest first. A note without a date is not listed.
last_update no YYYY-MM-DD. If it differs from date, the note shows "revised …".
summary yes Short text in the Notes list and in link previews.
standfirst no Italic intro under the article title.
topics no List. The first three show as "filed under"; all show at the end of the note.
neuron.type no pyramidal, purkinje, stellate or bipolar. Default pyramidal.
neuron.seed no Any whole number. Same seed and type always give the same cell; pick one no other note uses.
lenses no One line per lens: brain, learning, machine. The text is what shows in the Atlas. Leave a lens out to keep the note out of it.
math no Set to false to skip loading MathJax on a note without formulas.

Seeds in use: 11 (Plate I), 29 (Plate II). 47 is the "not written yet" ghost cell, and 3 and 17 draw the page backgrounds; avoid those.

Writing the body

Write normal Markdown. The note layout styles it automatically.

  • Sections: use ## for sections and ### for sub-sections (#### below that). Every ## heading becomes an entry in the contents sidebar. A leading number like ## 2. Rosenblatt's Mark I or ## 2) Bias in biology is split off and shown in the accent colour. Don't start the body with a # title; the layout already prints it.

  • Blank lines: leave an empty line between blocks (paragraphs, lists, headings, images, tables). Without one, Markdown glues them together: a sentence typed directly under a list becomes part of the last bullet.

  • Horizontal rules: always leave an empty line above ---. Directly under a line of text, Markdown turns --- into a heading instead.

  • Pull quotes: any > blockquote becomes a large italic quote. To add the small handwritten label above it, put an attribute line right after it:

    > If neurons are just circuits, why do we simulate them in software?
    {: data-label="the question"}
  • Figures: put an image on its own line and an italic caption on the next line. It becomes a numbered plate (fig. 1, fig. 2 …), with any leading "Figure:" removed from the caption:

    ![Perceptron Manual 1960](https://example.com/perceptron.jpg)
    *Figure: An illustration from the 1960 Perceptron Manual.*

    For images you own, put them in assets/images/ and link them as {{ site.baseurl }}/assets/images/name.jpg.

  • Term and description rows: use a definition list:

    Prior acceptance
    : holding a belief without re-deriving its justification every time
    
    Selective attention
    : overweighting certain evidence types based on pre-existing filters
  • Tables: normal Markdown tables. Wide tables scroll sideways on phones.

  • Math: $...$ inline and $$...$$ for display formulas (MathJax).

Checking a note before publishing (optional)

GitHub Pages builds everything for you, so this step is optional.

The simplest check needs nothing installed. Push the note to any branch other than main; the Build check workflow (.github/workflows/build-check.yml) builds the site with the same Jekyll that GitHub Pages uses and reports any error under the Actions tab. The built site can be downloaded there as the site artifact. Merge into main when it passes.

To preview locally you need Ruby and the github-pages gem, or Docker:

docker run --rm -it -p 4000:4000 -v "$PWD":/srv/jekyll -w /srv/jekyll ruby:3.3 bash -c "gem install github-pages webrick && jekyll serve --host 0.0.0.0 --baseurl /largelanguagemind"

Then open http://localhost:4000/largelanguagemind/.

If a pushed note doesn't appear, check the build under the repository's Actions tab. The usual causes are a missing date, a front matter line without a closing quote, or a tab character in the front matter.


Configuration

Atlas lenses

The three lenses live in _config.yml:

lenses:
  - { key: brain,    letter: a, title: Brain,    plate: cortex,     seed: 7,  sub: "..." }
  - { key: learning, letter: b, title: Learning, plate: tracts,     seed: 11, sub: "..." }
  - { key: machine,  letter: c, title: Machine,  plate: connectome, seed: 5,  sub: "..." }

key is the name notes use under lenses:. plate is the drawing (cortex, tracts or connectome), and seed varies it.

Domain

The site is served at https://theminddev.com/largelanguagemind. The domain itself belongs to the separate repository theMindDeveloper/theminddeveloper.github.io. That repository also redirects theminddev.com and theminddev.com/llm here. This is why _config.yml has url: "https://theminddev.com" and baseurl: "/largelanguagemind". If the repository is ever renamed, baseurl must change to match.

Brand assets

assets/brand/ holds the exported logo PNGs (favicon, apple-touch-icon, social preview). To re-export them, open assets/brand/logo-export.html in a browser from a local copy of the repository and click each button. It is not published on the site.

Design system

Colours, type, spacing and motion follow the design hand-off. The tokens are CSS custom properties at the top of assets/css/journal.css, in two sets: paper (default) and night (lights off). The visitor's choice is stored in localStorage under llm-theme. Article pages also offer a newspaper view (justified multi-column text with drop caps; headings, figures, tables and pull quotes span all columns), switched from the article header and stored under llm-read. It needs nothing extra in the Markdown: any note renders in both views. Visitors whose system asks for reduced motion get a hero that grows once and then stays still.


License

Copyright © 2025–2026 theMindDeveloper. All rights reserved. This is a personal brand: no part of this repository, including the notes, design, code, figures, name and logo, may be copied, reused, modified, redistributed or used to train AI systems without written permission. See LICENSE for the full terms.

Contact: theminddevlab@gmail.com

About

A personal research journal documenting my evolving understanding of human cognition, machine learning, and the foundations of intelligence.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages