it all stems from central two rules

1. assume the reader is going to read every word (if you have an llm assist you write, read every word before you paste)
2. be respectful of your reader’s time

in the bible there was a line about two great commandments. perhaps two rules is an irreducable.

if these rules are followed, than using llms can greatly enhance your writing, primarily by making better use of the reader’s time. how can we do this?

1. llms can be used to find objections and weakpoints in your posts, which can spur you to do more research and refine your idea. this saves the time that the reader might use to critique you and than rinse and repeat in a subsequent post. (this can also be used to steelman and so you shouldn’t steelman really bad positions)
2. the new search tools in llms are very good for finding sources specifically relevant to your piece (again, can also be used to steelman)
3. nomenclating – llms are exceptionally good at coming up with names for new ideas. especially if you don’t have a perfect knowledge of classical roots, or if you want to quickly find out about relevant terms from fields outside your own. llms can even be used in the proposed nomenclator
4. perhaps your topic has been written about before and you just don’t know it. asking an llm about relevant projects before can save you the time of doing work that someone else has already done.
   a good way (but not the only good way) to do this is try to come up with something that accomplishes three different points that hasn’t been done in that combination before (is it cheap? is it exquisite? does it run in the terminal? does it have fluid graphics? does it combine two disparate fields? is there a niche audience? a wide audience?)
5. summarizing – if you have duplicate statements of an idea that aren’t necessary (e.g. you have duplicates that aren’t there for emphasis; you do have the same idea expressed twice slightly differently in different parts of your piece), an llm can help you identify them and cut them. this too, is more respectful of the reader’s time

relevant: Please do not use AI to write for you

links

having an inline link in the opening {two paragraphs, 500 words} of your piece is not good practice because it could direct your reader away to another page. bad for retention

i use inline links when

• it is something like a quick reference page to a non-scholarly source like wikipedia for a definition for an uncommon term
• it is a “see also” for a tangentially related interesting piece
• when i am linking to another page in the same website

if the link is not to the same website, i use the attribute target="_blank" on the link to open it in a new tab. if it opens in the same tab you could lose your reader, which is also bad for retention

future versions of this site i plan to automatically color the links differently between same-site same-tab, new tab, or plain same-tab using css. would probably be a pandoc extension

i typically include one or two links to other posts on the same site to keep the reader engaged, and to show that IT IS ALL INTERCONNECTED

besides those usecases, for everything else for external links i use footnotes

indentations

i have a tendency to separate my headings from my content words, horizontally, using css. i try to use as much pure html as possible (using css gets you down a rabbit hole of customization and too many options too quickly), but the padding-right: 20px is too convenient.

i have made a few scripts to make notetaking easier.

central to my notetaking is a program i made called ENTS. a favorite way to use that program is to tag things on a “reading level” scale

+ reading level 
    - not read
    - skimmed
    - partially read
    - read
    - re-read

i use a shell script to make an html redirect file if i want to save a link but don’t want to save the file (if the file size is disproportionately big, if there are too many supplementary scripts, if the file is just not important enough to have an offline copy of)

#!/usr/bin/env bash

if [ $# -ne 2 ]; then
    echo "Usage: $0 <filename> <url>"
    exit 1
fi

FILENAME="$1"
URL="$2"

cat > "$FILENAME" << EOF
<!DOCTYPE html>
<html>
  <head>
    <meta http-equiv="refresh" content="0;url=$URL">
  </head>
</html>
EOF

echo "Created redirect file: $FILENAME"

then i can have the redirect file on my computer like it’s an actual file, and i can tag it with whatever tags i need to with ents

while reading i put things in an xml schema called EXRPT. i have haskell program which is not completely done yet which takes the xml and renders it to a nice-looking html rendering. i’ll publish the program later

During previous jobs I had some informal engineering training and I also previously ran a successful clothing startup.

But why do I feel the need to qualify and legitimize myself, I shouldn’t, really.

Currently I write TDSs (technical design specifications), white papers, proposals, whatever they are called in Markdown with Latex. I convert them to HTML with pandoc and publish them on the internet using an HSSG (hybrid static-site generator) I made called sculblog.

I generally write these with a startup context in mind. Prominent parts I include are:

• reason why making
• list of forces
• potential manufacturing techniques
• bill of materials

stylistic

it is unconventional to do this in prose writing but using curly braces {} to use sets of things helps remove a lot of clutter words in complex grammar structures (“in”, “or”, “with”) and it makes reading and writing complex sentences a lot easier. should probably used not too frequently because visually it is a bit jarring (because bread tastes better than key ). i did this once in this piece here

latex

whenever i use latex in pandoc, i use the --mathml option, which makes svg-like latex in the document, instead of embedding images or linking to an external javascript renderer (like mathjax). this is so if someone wants to save the document offline, it doesn’t have missing references

markup

i write in markdown. (i’ve switched between using obsidian and vim as an editor). the recent oversaturation of markdown llm outputs has made markdown output a little unsightly, which is a shame because it’s an extremely useful tool, so a few stylistic cleanups are needed

i used bulleted lists sparingly, and only when it would read a lot easier than reading it in a paragraph and it is in the middle of a flow. usually for a list of complex ideas. otherwise if it’s a list of simple things i keep it in a sentence/paragraph

i do not use those numbered lists where the first central term is bolded and then there is a dash and a description after it. who started that

for subsection headers, use bold text instead of a header element. the same font size makes it flow better

rarely do i go to a heading level below h2

llm output is probably formatted for the least common denominator, anyways.