keypnchr

Field Note · June 23, 2026

2026-06-23T00:00:00-05:00

"MarkItDown is a lightweight Python utility [from Microsoft] for converting various files to Markdown for use with LLMs and related text analysis pipelines."

github.com

"The Coming Loop"

lucumr.pocoo.org

"Vulnerability Reports Are Not Special Anymore"

filippo.io

Field Note · June 8, 2026

2026-06-08T00:00:00-05:00

Apple Core AI. Yes.

apple.com

A case for why every lab should have explicit AI-use guidelines.

substack.com

"AI Cannot Afford To Slow Down — It Needs $3 Trillion Or More In Revenue By End Of 2030 To Sustain Its Existence"

wheresyoured.at

Field Note · May 7, 2026

2026-05-07T00:00:00-05:00

SQLite recommended as archiving format by LOC sqlite.org

LOC-recommended data archiving formats loc.gov

Field Note · April 7, 2026

2026-04-07T00:00:00-05:00

"Apple built an LLM into your Mac. apfel gives it a front door." apfel.franzai.com

Field Note · April 5, 2026

2026-04-05T00:00:00-05:00

Caveman for Claude: "75% less word. Brain still big." github.edu/

Field Note · March 9, 2026

2026-03-09T00:00:00-05:00

"Shock! Shock! I learned yesterday that an open problem I'd been working on for several weeks had just been solved by Claude Opus 4.6 - Anthropic's hybrid reasoning model that had been released three weeks earlier! It seems that I'll have to revise my opinions about "generative AI" one of these days. What a joy it is to learn not only that my conjecture has a nice solution but also to celebrate this dramatic advance in automatic deduction and creative problem solving." — Donald Knuth stanford.edu/

A LaTeX-Style Writing Template for iA Writer

2026-03-06T00:00:00-06:00

Few things are as immediately recognizable as an academic document typeset in LaTeX: the Computer Modern font family, the justified text with first-line paragraph indentation, the clean hierarchy of section headings. These produce a look that Microsoft Word and its clones rarely replicate.

iA Writer is a distraction-free Markdown editor for those who prefer to work in plain text. iA Writer’s template system makes it possible to bring the typographic conventions of LaTeX into a Markdown workflow without touching a .tex file, or using an additional tool such as pandoc.

This template aims to do exactly that. It uses the Computer Modern Unicode (CMU) font family — the same fonts LaTeX loads by default — and applies the typographic rules of the standard article class: justified body text with first-line paragraph indentation, a clean three-level heading hierarchy, monospace code blocks in CMU Typewriter, and table styling in the spirit of the booktabs package. The result is a PDF that pays homage to those produced by LaTeX itself.

What the Template Includes

Download Template

The download contains four files:

LaTeX-Article.iatemplate — the iA Writer template bundle, ready to install
LaTeX-Article-Example.md — a sample document demonstrating every supported feature, including the title block, abstract, headings, lists, code blocks, tables, blockquotes, and footnotes
LaTeX-Article-Example.pdf — a rendered PDF of the example document, so you can see the output before installing
README.md — an overview document with installation and user instructions.

Who It Is For

The template is well suited to anyone who writes documents intended to look like conventional scholarly work — working papers, technical reports, draft manuscripts, course materials, or internal research memos. It is not a replacement for LaTeX in contexts that require precise equation typesetting or journal-specific formatting, but for prose-heavy documents where the goal is clean, credible presentation, it produces excellent results with ease.

How It Works

Installation takes a single double-click. Once installed, the template appears in iA Writer’s template chooser in the Preview panel. Selecting it applies the styling immediately; exporting to PDF preserves it.

Title Block

The title block convention is straightforward. Wrapping the opening lines of a document in a <div class="title-block"> element — with the paper title as a first-level Markdown heading followed by plain-text author and date lines — produces a centered, ruled title block in the structure of LaTeX’s \maketitle output. The rest of the document is written in standard Markdown.

All styling is contained in a single well-commented HTML file inside the template bundle, making it straightforward to adjust font size, line spacing, margins, or link color to suit a particular use case.

Download

The template is free to use and modify.

Download LaTeX Article for iA Writer (ZIP archive, includes template, example document, and example PDF)

The README included in the archive covers installation, the title block convention, section numbering, footnote syntax, and instructions for offline font use for those who work without a consistent internet connection.

Field Note · February 25, 2026

2026-02-25T00:00:00-06:00

Father of PageMaker, and coiner of "desktop publishing," dies. geekwire.com/

"Leading AIs from OpenAI, Anthropic and Google opted to use nuclear weapons in simulated war games in 95 per cent of cases." newscientist.com/

From a matplotlib maintainer: "An AI agent of unknown ownership autonomously wrote and published a personalized hit piece about me after I rejected its code, attempting to damage my reputation and shame me into accepting its changes into a mainstream python library." theshamblog.com/

"New in Claude Code: Remote Control. Kick off a task in your terminal and pick it up from your phone while you take a walk or join a meeting. Claude keeps running on your machine, and you can control the session from the Claude app or http://claude.ai/code" claude.com/

Field Note · February 24, 2026

2026-02-24T00:00:00-06:00

"We have identified industrial-scale campaigns by three AI laboratories—DeepSeek, Moonshot, and MiniMax—to illicitly extract Claude’s capabilities to improve their own models" anthropic.com/

Claude Code can now modernize COBOL code claude.com/

Field Note · February 18, 2026

2026-02-18T00:00:00-06:00

27-year-old Apple iBooks can still download official updates from Apple. reddit.com/

Field Note · February 12, 2026

2026-02-12T00:00:00-06:00

"The Most Expensive Party Balloon in History": on El Paso's airport shutdown. https://log.jasongodfrey.info/

Introducing GPT‑5.3‑Codex‑Spark openai.com/

Field Note · January 27, 2026

2026-01-27T00:00:00-06:00

"Addressing the risks of AI will require a mix of voluntary actions taken by companies (and private third-party actors) and actions taken by governments that bind everyone." darioamodei.com/

Field Note · December 10, 2025

2025-12-10T00:00:00-06:00

"Vibe coding actually works. It creates robust, complex systems that work. You can tell yourself (as I did) that it can’t possibly do that, but you are wrong. You can then tell yourself (as I did) that it’s good as a kind of alternative search engine for coding problems, but not much else. You are also wrong about that." stephenramsay.net

Red River Fall Retreat 2025

2025-10-18T00:00:00-05:00

.jpg

Click and drag to explore the 360° view

Field Note · September 22, 2025

2025-09-22T00:00:00-05:00

Cormac McCarthy notes on writing for science: gwern.net

Field Note · September 10, 2025

2025-09-10T00:00:00-05:00

UCLA digitized Hearst newsreels from 1929 to 1967 newsreels.net

Teen-agers in their bedrooms, before the age of selfies: newyorker.com

Field Note · September 8, 2025

2025-09-08T00:00:00-05:00

Better late than never — iPad native Instagram app available instagram.com

White paper: Why Language Models Hallucinate openai.com

Field Note · September 4, 2025

2025-09-04T00:00:00-05:00

Not new, but worth not missing: "Lawrence Livermore National Laboratory expands Claude for Enterprise use to empower scientists and researchers" anthropic.com

Local LLMs on Apple Silicon with Ollama + WebUI

2025-08-08T00:00:00-05:00

If you are running a Mac with at least an M1 processor (an M4 with 24GB of RAM handles these surprisingly well), and you were curious about running large language models locally on your machine, I have put together a guide to take you from start to finish on getting that up and running.

Open your terminal to a command prompt and get started below…

Prerequisites

Mac running Apple Silicon (at least M1)

Homebrew (https://brew.sh ). If you don’t have it:

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

1. Install Ollama

Ollama runs models locally and exposes an API on http://localhost:11434.

brew update
brew install ollama
brew services start ollama

The final command starts a background service listening on port 11434.

Verify it’s up and running:

ollama --version
curl http://localhost:11434/api/tags

The last command should return some JSON. If you don’t see an error, you should be good.

2. Pull some models

Download the model weights you want. Here are some examples:

## OpenAI Model
ollama pull gpt-oss
## Meta Model
ollama pull llama3
## Google Model
ollama pull gemma
## Mistral Model
ollama pull mistral:7b
## Multimodal Model (vision recognition and text)
ollama pull llava:7b

List your downloaded models:

ollama list

To remove an installed model:

ollama rm <model-name>

3. Install WebUI

WebUI provides a graphical browser interface to interact with the local Ollama server. Install with:

brew install python@3.12
brew install pipx
pipx ensurepath
pipx install --python /opt/homebrew/opt/python@3.12/bin/python3.12 open-webui

Run the following commands in your terminal to ensure the path needed for WebUI remains available (only needs to be done once):

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zprofile
source ~/.zprofile

4. Run a Model and Interact using WebUI

Choose a model to run (OpenAI’s gpt-oss in this example):

ollama run gpt-oss

Now, run WebUI. You will be prompted the first time you run it to set up an admin account:

open-webui serve

Open your browser to http://localhost:8080 (first launch will walk you through creating an admin user).

5. Optional: add aliases to your zsh terminal profile for easier access.

Edit your .zprofile:

nano ~/.zprofile

Add the following lines to your ~/.zprofile based on which LLMs you decide to install (be sure to add in the WebUI line at the bottom of the list; you can change Safari to another browser if you prefer):

alias gpt='ollama run gpt-oss'
alias llama='ollama run llama3'
alias gemma='ollama run gemma'
alias mistral='ollama run mistral:7b'
alias llava='ollama run llava:7b'

alias webui="open-webui serve & sleep 5 && open -a Safari http://localhost:8080"

Hit Ctrl+X to exit. Choose (Y)es to save the file, then enter to save as same file name.

After saving, reload the ZSH profile (you only have to do this once; each time you open a new terminal it will already be there):

source ~/.zprofile

Now you can quickly launch a model in the terminal, e.g.:

gpt

…and launch WebUI with:

webui

SPECIAL NOTE: there is a 5 second delay after running the WebUI command before it attempts to open the page in a browser (http://localhost:8080). If it has not finished starting in the terminal the page won’t load yet, but you can refresh the page when it is finished to begin.

Ollama via Homebrew already runs as a service:

brew services start ollama
brew services stop ollama
brew services restart ollama

7. Troubleshooting

Reset/restart services

brew services restart ollama

What’s using a port (11434 or 8080)?

lsof -i :11434
lsof -i :8080

8. Uninstallation

Uninstall Ollama (Homebrew):

brew services stop ollama
brew uninstall ollama
rm -rf ~/.ollama

Uninstall Open WebUI (pipx):

pipx uninstall --python /opt/homebrew/opt/python@3.12/bin/python3.12 open-webui

9. Security & privacy notes

Everything runs locally by default; models and prompts don’t leave your machine unless you enable remote features.
Treat downloaded models like large binaries; verify sources before pulling community models.

Field Note · August 6, 2025

2025-08-06T00:00:00-05:00

OpenAI releases two open-weight models.

"The gpt-oss-20b model delivers similar results to OpenAI o3‑mini on common benchmarks and can run on edge devices with just 16 GB of memory, making it ideal for on-device use cases, local inference, or rapid iteration without costly infrastructure" openai.com